Start with the question, not the chart type
Most confusing charts begin with a tool’s menu of options rather than a question. Before you pick anything, write down in one sentence what you want the reader to take away: “Sales rose each month this quarter,” “Most of our requests come from two categories,” or “Costs are split roughly evenly between three items.” The chart is then a way of showing that sentence, and if you cannot write the sentence, the chart is unlikely to help.
It also helps to decide who the reader is. A colleague who knows the data can handle more detail than a client seeing it for the first time.
Matching chart types to questions
A few chart types cover nearly every everyday need.
Line charts show change over time. Because the eye follows the line from left to right, they suit trends such as weekly orders or monthly expenses. Use them when the horizontal axis is a sequence of dates.
Bar or column charts compare amounts across categories, such as hours by project or requests by type. Sorting the bars from largest to smallest, unless the categories have a natural order like months, makes comparison easier.
Pie charts show parts of a whole, but they are hard to read when there are more than a few slices or when slices are similar in size, because people judge angles less accurately than lengths. Many writers prefer a bar chart for the same data. If you do use a pie, keep to a handful of slices and label them directly.
Scatter plots show whether two numeric measures seem to move together, such as hours worked and tasks completed. They are less familiar to many readers, so they usually need a sentence explaining what to look for. A visible pattern also does not prove that one thing causes the other.
Tables are sometimes better than any chart. If readers need exact values, or you have only three or four numbers, a small table is often clearer than a graphic.
Labels that do real work
A chart should be understandable without the surrounding paragraph. A few labels make that possible.
- A title that states the point or the content. “Support requests by week” describes; “Support requests rose after the March change” claims something. Either is fine as long as it is accurate and the data supports it.
- Axis labels with units. Say whether a number is dollars, hours, percent or count, and whether it is thousands.
- Direct labels over legends. When possible, label a line or bar next to the thing itself rather than forcing the reader to match colors against a key.
- A source and date range. A short note such as “Data: our own task log, January to June” lets readers judge what they are seeing.
Axes: the quiet source of confusion
Bar charts should start at zero. The length of a bar is the message, and cutting off the bottom exaggerates differences. If one bar is 52 and another is 50, starting the axis at 49 makes the first look several times taller than the second, though the real gap is 2 units, or about 4 percent of 50. (These figures are hypothetical, used only to show the arithmetic.)
Line charts have more flexibility, because the reader is looking at slope rather than length, but a heavily zoomed axis can still make small wiggles look dramatic. If you do zoom in, say so on the chart.
Keep scales consistent when charts sit side by side. If two charts use different vertical ranges, a reader may compare them by eye and reach the wrong conclusion. Dual axes, with one measure on the left and another on the right, are easy to misread, so use them rarely and label each axis clearly.
Use color with restraint
Color is best used to direct attention. A common approach is to draw most elements in a neutral gray and use one strong color for the item you want noticed. A chart in six bright colors gives the reader no hint about where to look.
Some readers have color-vision differences, and charts are often printed in black and white. Do not rely on red versus green alone to carry meaning. Add direct labels, patterns or different line styles. Choosing tools with good defaults is part of software accessibility, and it applies to charts as much as to menus.
Remove what does not help
Default chart settings often add heavy gridlines, borders, shadows and three-dimensional effects. Three-dimensional bars and pies in particular distort how large the shapes look. A good rule is to delete each element and see whether the chart is still understood; if so, leave it out. Light gridlines or none at all, a clean background and a modest font size usually serve better than decoration.
Limit how many series you show. A line chart with eight lines tends to turn into a tangle. Either highlight one line and gray the rest, or split the data into several small charts that share the same scale.
Be honest about the data behind it
A clear chart is not the same as a persuasive one. A few cautions apply:
- Show the full relevant time range rather than a convenient slice that flatters the story.
- Mention when the sample is small. A percentage based on five responses can swing widely.
- Keep averages in context. An average hides spread, so a note about the range, or a chart of the individual values, can add honesty.
- If data was cleaned, filtered or had missing values removed, say so briefly. Messy inputs lead to misleading outputs, which is why data cleaning comes first.
A quick pre-share check
Before sending a chart, try showing it to someone for five seconds and asking what they think it says. If the answer matches your one-sentence message, it is working. If not, adjust the type, simplify the labels or remove an element and test again. Also check that the numbers in the chart match the source table, since a stale chart pasted into a report is a common error. Many team dashboards are made of several small charts, so the same principles apply when you combine them on one page.