DataslopeWhat Is Data Visualization, Really?
A careful definition, with examples, of what counts as a "data visualization" and what doesn't
You have read several chapters of this course already and you have seen many charts. We have not yet asked the most basic question: what is data visualization?
It sounds like a silly question. Everyone "knows" what a chart is. But spending five minutes on a careful definition will save you months of confused work later. Many of the worst visualizations in the world were made by people who never paused to ask what they were actually building.
A working definition
Here is the definition we will use for the rest of this course:
Data visualization is the practice of encoding information into a visual representation so that a human visual system can recognize patterns and answer questions that would be difficult to answer from the raw data alone.
Three words in that sentence are doing all the work: encoding, visual, and questions.
Encoding
A visualization is a mapping from data to visual properties. Numbers become positions. Categories become colors. Time becomes a horizontal axis. The map is called an encoding.
The encoding is the design decision. There is no "neutral" or "default" encoding. Every chart you make is the result of choices about how to translate columns of data into ink (or pixels).
Visual
Visualization is specifically about the visual system — the parts of the brain that process light, edges, motion, color, and shape. We are using this biological hardware because it is extraordinarily fast and parallel. A human can detect that one bar is taller than another in roughly 200 milliseconds, without conscious thought. Compare that to reading two numbers and doing a mental subtraction, which takes several seconds.
This means we are taking advantage of built-in capabilities. A good chart works with the visual system; a bad chart works against it.
Questions
A visualization exists to answer a question. The question may be vague ("what's going on in this dataset?") or precise ("did sales increase last quarter compared to the same quarter last year?"), but there is a question. A chart without a question is decoration.
This is the cleanest test for a visualization:
What question does this chart answer?
If you can't say in one sentence, the chart probably doesn't earn its space.
What is not a data visualization?
Sharpening the definition helps. Here are some things that look like visualization but aren't, or aren't quite:
- A photograph is not a data visualization. It is a visual recording, but it doesn't encode abstract data — it captures what was there.
- A logo or an icon is not a data visualization. It is a pictogram or branding element.
- A formatted table with colored cells is barely a data visualization. Cell color is an encoding, but the underlying structure is still rows of numbers, not a visual representation. It's a useful borderline case.
- A diagram explaining a process (like a flowchart or a network diagram) is a visualization, but a different kind — it visualizes relationships, not data values. We sometimes call these "diagrammatic" or "schematic" visualizations.
- A 3-D rendering of a building is a visualization of the building, not of data about the building. The distinction is whether the visual is abstracting numerical information or representing a physical thing.
For this course, we mean statistical / analytic visualization: encoding measurements into visual properties of a 2-D picture.
A live example: same data, different visualizations
Watch what changes when the same five numbers are shown in three different ways. Each is a legitimate visualization — and each answers slightly different questions.
The data is identical in all three charts. But:
- The bar chart invites you to compare heights. Apple is ~4× Date.
- The pie chart invites you to compare shares of the whole. Apple is "about a third" of total sales.
- The dot chart is the same idea as the bar but with less visual weight per data point — useful when you have many categories.
The data does not change. The encoding changes, and so the question being answered changes. This is the whole game.
Why this is worth dwelling on
If you take only one thing from this page, take this:
A visualization is a deliberately-designed encoding of data into a visual representation, made to answer a specific question for a specific audience.
That sentence — boring as it is — will protect you from most of the mistakes beginners make. Whenever you find yourself making a chart, stop and ask:
- What is the data? (What are the rows and columns?)
- What is the question? (What am I trying to answer?)
- What is the audience? (Who will read this?)
- What is the encoding? (Which columns become which visual properties?)
When all four are clear, the chart almost makes itself.
Check your understanding
According to the definition used in this course, which of these is the clearest example of a data visualization?
A photograph of a sunset.
A company logo.
A bar chart of monthly revenue for a small business.
A 3-D rendering of a planned building.
What does it mean to say a chart "encodes" data?
It applies a password to the chart file.
It uses statistical formulas internally.
It maps columns of data to visual properties such as position, length, color, size, or shape.
It compresses the data.
A colleague hands you a beautifully styled chart with no clear takeaway. You ask "what question does this answer?" and they shrug. What is the most likely problem?
The chart is technically broken.
The data is wrong.
The chart was created without a clear question in mind — so even if it's pretty, it isn't really doing anything as a visualization.
The chart needs more colors.
Why is it important that data visualization works with the human visual system?
Because computers can't draw without humans.
Because charts are required to be pretty.
Because the visual system can recognize patterns (relative size, position, color) in milliseconds — much faster than reading raw numbers.
Because eyes are the only sensors humans have.