Chart Types: When to Use Which (and When You're Wasting a Pie Chart)

The default chart for 'how does X compare across categories'. Revenue by region, sign-ups by channel, units by product. Each bar is a category, the height (or length, if horizontal) is the value.

Use a horizontal bar chart when you have many categories or long category names — vertical labels become illegible past about ten bars. Sort by value, not alphabetically, unless there's a meaningful order (like months or grades). The eye reads sorted bars dramatically faster.

Time on the X axis, your metric on the Y axis, dots connected with a line. The line implies continuity — that the value moves smoothly between measurement points — so reach for line charts when that's a fair assumption (revenue, active users, temperature) and avoid them when it isn't (winners of an election by year — those are discrete events, use bars).

Keep the number of lines small. Three or four is a comfortable maximum. Past that, the chart becomes spaghetti and nobody can read which line is which. If you genuinely have ten time series, use small multiples (one mini-chart per series, laid out in a grid) instead of cramming them onto one panel.

Each dot is one observation, plotted by its X and Y values. A scatter plot tells you whether two numeric variables move together, against each other, or unrelated.

Reach for it when you suspect a relationship and want to see it. Spend per customer vs lifetime value. Page load time vs conversion. Hours of sleep vs energy level. The shape of the cloud is the answer: tight upward = positive correlation, scattered cloud = no relationship, downward = inverse.

A common upgrade: add a trend line (linear or LOESS) so the eye doesn't have to do the regression itself. Just don't extrapolate beyond the data you actually have — the trend line is descriptive, not predictive, unless you've explicitly fit a model.

Pie charts work for one specific job: showing parts of a whole when there are very few slices (two or three) and the difference between them is dramatic. 70% / 30% as a pie is fine. 18% / 16% / 15% / 14% / 12% / 11% / 14% as a pie is a war crime.

The human eye is bad at comparing angles and curved areas. It's much better at comparing the lengths of straight bars. So 95% of pie charts should be horizontal bar charts. The rare exception is when 'this big slice and that small slice' is the entire story you're telling and you want the visual punch.

Avoid 3D pie charts, exploded slices, and any decoration. They distort the proportions. If you find yourself reaching for them, reach for a different chart type instead.

A bar chart where each bar is sub-divided into segments. Useful when you want to show both the total per category AND the composition. Revenue by month, broken down by product line. Sign-ups by week, broken down by channel.

Keep the number of stacked segments small (3-5 max) and order them consistently across bars. The biggest pitfall: stacked bars make it hard to compare individual segments across categories — you can compare the segment at the bottom (it shares a baseline) but not the ones above it (they float).

When comparing segments matters more than seeing totals, switch to a grouped bar chart (segments side-by-side instead of stacked) or 100% stacked bars (every bar is normalised to 100%, so you're comparing share, not absolute).

Box plots show the spread of a numeric variable across categories: median, quartiles, outliers. The box covers the middle 50% of the data, the line in the middle is the median, the whiskers extend to the typical range, dots beyond the whiskers are outliers.

Use them when you want to compare distributions across multiple groups — salary by department, response time by server, test scores by school. They condense a lot of information into a small space, which is also the downside: people who haven't seen one before need a quick explanation.

Violin plots are box plots' richer cousin. Same job — comparing distributions across groups — but instead of a box, each group gets a smoothed shape that shows the actual probability density. A violin plot reveals when a distribution is bimodal (two humps), skewed, or unusually peaked — features a box plot hides.

Use them when the shape of the distribution matters and your audience can read them. If you're presenting to executives who've never seen one, a box plot is more readable. If you're a data team comparing a model's predictions across cohorts, the violin plot's extra detail is worth the cognitive load.

**3D anything.** Three-dimensional bar charts and pies distort the proportions. Almost no use case justifies them.

**Radar / spider charts.** Look impressive, almost impossible to read accurately. The closed shape implies a meaningful 'area' that often isn't.

**Donut charts with annotations.** A pie with a hole. The hole adds nothing; the labels around it usually do all the work, meaning the chart itself is decoration.

**Word clouds.** Show neither frequency clearly nor relationship — they show big-text-vs-small-text and that's it. If you need to convey 'these terms came up most', a sorted bar chart does it in less space and more accurately.

None of these are illegal. They're just usually the wrong tool, and if you reach for them, ask yourself what specific question they're answering that a simpler chart wouldn't answer better.

A dashboard rarely needs more than 3-5 chart types. KPI tiles for the headline numbers. A line chart or two for trend. One bar chart for breakdown. Maybe a table for the long tail.

The biggest mistake we see is dashboards that try to use every chart type at once — like a designer's sample file. The user's eye doesn't know where to land. Pick the question the dashboard answers, pick the 4-5 visuals that answer it, drop the rest.

We built AnalityQa partly because picking the right chart type is its own skill, separate from the data work. You ask the question; we pick the chart that fits. You can override it (every chart type is one click away in the editor) but the default usually lands close to the right answer.