Charts that tell the truth: choosing the right chart and avoiding misleading ones

A chart is a claim. When you put one on a slide, you are telling the reader what the numbers mean before they have read a single figure. That is why charts persuade so well, and why a careless chart can mislead a board just as easily as a deliberate one.

The good news is that most honest charting comes down to a few rules. UK government statisticians have written many of them down, and this guide draws on the Government Analysis Function and Office for National Statistics (ONS) guidance, plus the well-documented ways that graphs go wrong.

Start with the message, then pick the chart

The Analysis Function’s guidance on charts begins with a simple test: if you cannot state the message in a few sentences, simplify the chart or split it up. Write the message first, then choose the form that shows that relationship.

That last row matters. The same guidance says a well-structured table can often be as powerful as a chart, and is the right choice when people need exact values rather than an overall impression.

The zero rule: bars yes, lines not always

The most common way charts mislead is the truncated axis, where the vertical axis starts above zero so small differences look large. The Wikipedia article on misleading graphs notes that readers still overestimate differences even when they are told the axis is truncated, and that Excel often truncates by default when values sit in a narrow range.

The ONS position is clear and nuanced. In its blog post on bar charts, it says bar charts should always include zero, because readers compare the length of the bars rather than reading the axis. Cut the axis and a small gap becomes a big one. The ONS service manual on axes and gridlines applies the same rule to filled area charts.

Line charts are different. They encode values by position, not length, so a cropped axis can be acceptable and often shows the trend more clearly. The ONS blog post on line charts calls this an editorial decision and gives cautions: check how close to zero the data sits, consider whether plotting the change itself would be clearer, and mark the break so the reader knows. If you need a non-zero range to compare categories with small differences, ONS suggests a dot plot instead of a cut-off bar chart.

Other ways charts mislead

• Dual axes. Two series on two different vertical scales can make one appear to overtake the other when it has not. Both the ONS service manual and the Analysis Function advise against dual-axis charts and suggest two separate charts instead.
• 3D effects. Perspective distorts the size of bars and slices. The Analysis Function lists 3D shapes, shadows and heavy gridlines among things to avoid.
• Pie charts with many similar slices. People struggle to compare angles. The guidance limits pies to five categories or fewer that add up to a meaningful whole, with one category dominating.
• Scaled pictures. Enlarging an icon in two dimensions makes a doubling look like a quadrupling.
• Selective time windows. Choosing the start date shapes the story. ONS points out that a longer time span can give a very different picture.
• Unlabelled log scales. They can make very different values look similar. ONS recommends avoiding non-linear scales unless there is no clearer option.

Titles, labels and colour

The Analysis Function recommends two titles where you can: a headline that states the message, such as “Returns fell after the new size guide”, and a formal subtitle that says what the data is, where it covers and the time period. A headline title helps the reader, but it also commits you. If the chart does not clearly show what the title claims, change one of them.

Label lines directly instead of relying on a legend, and keep to about four lines on one chart. Use round-number gridline intervals, keep gridlines light grey, and label each gridline with its value. Do not rely on colour alone to tell series apart, because some readers will not see the difference. The guidance asks for charts to meet WCAG 2.2 at levels A and AA and to come with a text alternative: a short description of the message, a data table, or both.

A worked example (illustration only)

An online clothing brand’s return rate was 11.2% in Q1 and 10.6% in Q2. The marketing lead makes a bar chart in Excel, which sets the axis from 10.2% to 11.4%. The Q2 bar looks half the height of Q1, and the slide reads “Returns halved”.

A better version:

1. Start the bar axis at zero. The two bars now look almost the same height, which is honest: the rate fell by 0.6 percentage points.
2. If the change is the point, show it directly. A line chart of the monthly return rate over two years, with a cropped and clearly marked axis, shows whether Q2 is a real shift or normal variation.
3. Rewrite the headline to match: “Return rate down 0.6 points in Q2”, with a subtitle naming the data source and period.
4. Add a one-sentence text description for anyone who cannot see the chart.

Common mistakes and a quick self-check

Before any chart goes into a report, ask:

• Does a bar or area chart start at zero?
• If a line chart axis is cropped, is the break obvious?
• Could someone misread a second axis, a 3D effect or a pie?
• Does the time window include enough history to judge the change?
• Does the title say only what the chart shows?
• Would a table serve the reader better?

The most common failure is accepting Excel’s defaults. Excel picks axis ranges automatically, and ONS specifically warns that automatic chart builders often choose ranges you would not choose yourself. Check the axis every time.

When to chart and when not to

Chart when the reader needs a pattern: a trend, a ranking, a gap, a relationship. Use a table when they need exact figures, or when there are only two or three numbers, where a sentence often does the job better than any picture. And when the honest version of a chart looks less dramatic than you hoped, publish the honest one. The reader will trust the next chart you show them.