How to choose a chart type — and the two ways charts lie

The chart type is the message. A bar says "compare these"; a line says "this changed"; a pie says "these are parts of one thing". Pick the wrong one and readers draw the wrong conclusion before reading a label — or the right conclusion from a chart that was built to mislead. This guide is the ten-minute version of what data-visualisation courses spend a week on; the chart maker applies the defaults.

Start with the question

Every chart answers one question. Which is biggest? (comparison). What happened over time? (trend). How is the whole divided? (composition). How are values spread? (distribution). Are two things related? (relationship). Name the question first and the chart type follows; a chart that tries to answer two questions answers neither. Then decide what one sentence the reader should say after looking — and title the chart with that sentence, not with "Sales by month".

The four everyday charts

• Bar chart — comparison across categories. Sort the bars unless the categories have a natural order; use horizontal bars when labels are long. Grouped bars compare two or three series; beyond that, use several small charts.
• Line chart — change over a continuous axis, almost always time. The horizontal axis must be a sequence; a line through unordered categories implies a trend that doesn't exist. Up to four or five lines are readable; more becomes spaghetti.
• Pie or donut — parts of one whole, where the whole is meaningful and the parts are few. Beyond five or six slices, or when slices are similar in size, eyes can't compare angles — use a sorted bar chart instead. Never a pie of things that don't sum to 100%.
• Table — when readers need exact values or will look up one number. A table with sparklines beats a cluttered chart. The markdown table generator makes one.

Histograms (distribution) and scatter plots (relationship) are the next two; the standard deviation calculator and the average calculator supply the numbers behind them.

The axis that must start at zero

A bar's length is its value, so the bar axis must start at zero — cut it at 90 and a 3% difference becomes a cliff, the single most common visual lie in business reporting. A line chart's axis may be zoomed, because a line shows change, not magnitude; but say so, and don't zoom so far that noise looks like a trend. Dual axes (two scales on one chart) let anyone make any two lines appear correlated by choosing the scales — avoid them, or plot the two as separate small charts.

The two ways charts lie

Beyond the axis: encoding lies, where the visual doesn't match the number — 3D pies whose front slices look bigger, bubbles scaled by diameter rather than area (a 2× value drawn 4× as large), pictograms that grow in two dimensions. And selection lies, where the chart is accurate and the data is cherry-picked — a time window that starts at the low point, a comparison that omits the inconvenient category, a percentage of a tiny base. The defence is the same for making and reading: ask what the axis starts at, what's been left out, and whether the shape you see would survive a different window.

Honest defaults for a chart in a hurry

1. One question, one chart, a title that states the finding.
2. Bars from zero, sorted. Lines only over time. Pies only for a few parts of one whole.
3. Label directly where possible instead of a legend; readers shouldn't have to look back and forth.
4. A palette that survives greyscale printing and colour blindness — check it with the colour-blindness simulator.
5. Source and date in small text at the bottom.

Data comes in from Excel to CSV or the CSV viewer; the chart goes out as a PNG for the slide.