Data visualization

Data visualization is the graphical representation of data so that patterns, comparisons and trends can be understood quickly. It works because human visual processing detects differences in position, length and direction far faster than it can read equivalent numbers in a table.

That advantage is also the risk. A chart makes a pattern feel obvious whether or not the pattern is real, which means visualization choices carry a responsibility that formatting choices do not.

What Is Data Visualization?

Data visualization encodes values as visual properties — position, length, angle, area, color, shape — so relationships between them can be perceived rather than calculated.

Not all encodings are equally effective, and this is a matter of perception rather than taste. People judge position along a common scale most accurately, then length, then angle, then area, then color intensity. This is why a bar chart communicates comparison better than a pie chart, and why bubble sizes are consistently misread — area is a weak encoding and most viewers under-estimate large differences in it.

The practical consequence is that chart selection should follow from what needs to be perceived. Choosing a visual because it looks distinctive usually means selecting a weaker encoding for the actual question.

Choosing the Right Chart

Comparison between categories — a bar chart, horizontal when labels are long. Ordered by value unless the category order carries meaning.

Change over time — a line chart. Use bars only when periods are few and discrete.

Composition — a stacked bar, or a treemap for many categories. Pie charts work only with two or three segments, and even then a bar usually reads better.

Relationship between two variables — a scatter plot, with a trend line only where a relationship is genuinely being claimed.

Distribution — a histogram or box plot. This is the most frequently skipped question, which is why averages so often conceal the thing that matters: a mean is identical for a tightly clustered population and a bimodal one.

Part of a whole over time — a stacked area chart, keeping in mind that only the bottom series is easy to read accurately.

Design Principles That Matter

One chart, one point. A visual carrying three findings usually communicates none of them. Split it.

Remove what does not carry information. Heavy gridlines, borders, backgrounds, 3D effects and drop shadows add visual load without adding meaning. 3D in particular actively distorts, because perspective changes apparent size.

Label directly where possible. Putting series names next to the lines beats forcing the reader to look up a legend and hold the mapping in memory.

Use color deliberately. Color should encode something — a category, a threshold, a highlight. If every series has a different color for no reason, color communicates nothing and adds noise.

Give the chart a conclusion, not a label. A title reading “Q3 revenue by region” describes the axes. One reading “North region drove the Q3 decline” tells the reader what they are looking at.

Applying these consistently across an estate is a design-system problem rather than a per-chart one, which is where data product design earns its place.

How Visualizations Mislead

Truncated axes. A bar chart starting at 80 rather than zero turns a 4% difference into an apparent doubling. Bars encode by length, so the baseline must be zero. Line charts measuring change over time are the legitimate exception.

Dual axes. Two series on different scales can be made to appear correlated or divergent almost at will by adjusting the ranges. Treat any dual-axis chart with suspicion.

Selective time windows. Starting the period at a convenient trough makes any subsequent series look like growth.

Aggregation that hides the signal. A stable average concealing two segments moving sharply in opposite directions.

Area and 3D encodings. Doubling a circle’s radius quadruples its area, which viewers read as roughly double. The distortion is systematic and predictable.

Most of these are unintentional. They happen because the analyst formed a conclusion first and then chose, without deliberate intent, the presentation that supported it.

Dashboard Design

A dashboard is not a collection of charts; it is an answer to a recurring question for a specific audience. Designing it starts with what decision the viewer makes and how often, not with which measures are available.

Useful structure puts the most important measure top-left, where attention lands first, and arranges everything else in the order the viewer reasons — overall position, then what changed, then where. Every figure needs a comparison point; a number without a target, prior period or benchmark cannot be judged.

Two things are consistently missing and consistently matter: a visible indicator of when the data was last refreshed, and a limit on scope. Dashboards that try to serve executives, managers and analysts at once serve none of them well, because those three need different levels of aggregation. Our business analytics teams design to the audience rather than to the available measures.

Accessibility

Around one in twelve men has some form of color vision deficiency, most commonly affecting red and green — the exact pair used for good and bad in most business dashboards.

The fixes are straightforward. Never rely on color alone: pair it with shape, label, position or pattern so the distinction survives without it. Use colorblind-safe palettes such as blue-orange rather than red-green. Maintain sufficient contrast between text and background. Ensure charts remain legible when a user zooms.

These are not concessions that compromise design. A chart that works without color is almost always clearer for everyone, because it cannot lean on color to do work that structure should be doing.

Frequently Asked Questions