Visualisation
Choosing the chart that answers the question, designing a dashboard people use, and telling the story that gets acted on.
A Chart Is an Argument
Every chart makes a claim. The choice of encoding — position, length, angle, colour, area — decides how accurately a reader can decode it, and human beings read some encodings far better than others.
Position on a common scale is the most accurate. Length is next. Angle and area are noticeably worse, which is the real reason pie charts are criticised: readers cannot compare wedges reliably. Colour intensity is the weakest of all, and should carry your least important variable.
Choosing a Chart
Start from the question, never from the chart library. The question dictates the form.
| The question | The chart | Watch out for |
|---|---|---|
| How has this changed over time? | Line | Truncated y-axis exaggerating the trend |
| How do these categories compare? | Bar | Bars must start at zero |
| How are these values distributed? | Histogram or box plot | Bin width changes the story |
| Do these two things move together? | Scatter | Overplotting hiding density |
| What is the composition? | Stacked bar | Only the bottom segment is easy to compare |
| Where is it concentrated? | Heatmap | Colour scale doing too much work |
Never use colour alone to carry meaning. Roughly one man in twelve has some form of colour vision deficiency, so pair colour with shape, label or position.
Dashboard Design
A dashboard answers a recurring question for a specific person. A dashboard that answers everything for everyone gets checked twice and then ignored.
Lead with the headline number, then the trend, then the breakdown. Someone should be able to glance for three seconds and know whether anything needs their attention — that is the whole design brief.
| Principle | In practice |
|---|---|
| One audience | Ask whose decision this serves; build for them |
| Context beside every number | "1,204" means nothing; "1,204, up 8% on last week" does |
| Same period everywhere | Mixed date ranges on one screen create false comparisons |
| Fewer panels | Every extra chart lowers the attention paid to the rest |
| State the definition | Two teams will otherwise mean two things by "active user" |
Communicating Results
Lead with the conclusion. Analysts are trained to build up to a finding the way the investigation happened; decision-makers need the answer first and the method only if they challenge it.
Structure that works: what we found, what it means, what to do, then the method in an appendix. Say what would change your mind, and be explicit about uncertainty — an analyst who states confidence intervals is trusted more, not less.
Interview Questions
Why must bar charts start at zero when line charts need not?
A bar encodes value as length, so truncating the axis distorts the ratio the reader perceives. A line encodes change as slope, and the reader is comparing direction rather than magnitude.
What is wrong with pie charts?
They ask readers to compare angles and areas, which people judge far less accurately than position or length. A bar chart answers the same question better. They are tolerable for two or three parts of a whole.
How do you design a dashboard people actually use?
Pick one audience and one recurring decision. Headline number, then trend, then breakdown. Context beside every figure, consistent date ranges, and fewer panels than you want.
How do you present a result nobody wants to hear?
Lead with the finding, show the guardrails you checked, state the uncertainty honestly, and say what evidence would change the conclusion. Burying it costs your credibility the next time.
Why not rely on colour alone?
Colour vision deficiency affects roughly 8% of men, and colour is also the least accurately decoded encoding for everyone. Pair it with shape, label or position.