Data visualisation that tells the truth: charts for development reports

Development reports are full of numbers: enrolment rates, households reached, budgets spent, targets met. Good data visualisation for NGOs, UN agencies and ministries turns those numbers into something a busy reader can understand in a few seconds. Poor visualisation does the opposite, or worse, gives a false impression.

Before opening any software, write one sentence that says what the chart should show. For example: "Enrolment rose in every district, but fastest in the Eastern Province." That sentence can often become the title. The UNECE guide to presenting statistics says a chart title should give a clear idea of what the chart is about, and describes titles that state the main pattern as well as titles that state what, where and when.¹

Choosing the right chart starts with the question the reader is asking. Is it a comparison between places, a change over time, a ranking or a share of a whole? The Financial Times Visual Vocabulary, reproduced in the European data portal's data visualisation guide, is a set of chart types organised by what aspects of your data each type can show best.² It is a useful poster to keep next to anyone who builds charts.

For most development reports, a few chart types do almost all the work. The UNECE guide calls the bar chart the simplest type of chart to draw and read, and describes the line chart as an effective tool for showing trends over time.¹ It also advises sorting bars from smallest to largest so they are easier to compare, and notes that text is easier to read when bars are horizontal.¹

• Comparing districts, programmes or countries: a sorted horizontal bar chart
• Showing change over several years: a line chart
• Showing one share of a whole: a single bar or a simple percentage, often clearer than a pie
• Showing where something happens: a map, but only when location is the point

Pie charts are popular in annual reports, but they are hard to read when slices are similar in size. The UNECE guide says a pie chart can show the percentage distribution of one variable, but only for a small number of categories, usually not more than six, and adds that many statisticians do not recommend this type of chart.¹

If you need to show how a budget is split across eight budget lines, a sorted bar chart with the percentages written on each bar will almost always be easier to read. Keep pie charts for a single, simple message, such as one large share against the rest.

Most misleading charts are not made on purpose. They come from software defaults or a wish to make a small change look bigger. The most common problem is a bar or column chart whose axis does not start at zero. Datawrapper, a tool widely used by newsrooms, treats truncated y-axes in column and bar charts as deceptive and misleading, and does not let users start them above zero.³ The UK Office for National Statistics reaches the same conclusion: bar charts should always include zero, because readers compare the length of the bars.⁴

Line charts are different. Datawrapper notes that, like dot plots, line charts can but do not need to start at zero.³ If a line chart's scale does not start at zero, the UNECE guide recommends making that visible, for example with a zigzag line or a break in the axis.¹

Two other habits deserve caution. The UNECE guide says three-dimensional charts rarely add value and often confuse readers, and that dual y-axes have the potential to cause confusion.¹ If two measures really belong together, two small charts placed side by side, sharing the same time axis, are usually easier to read than one chart with two scales.

A chart without context invites the wrong conclusion. Say what the numbers count, over what period and for which population. When groups differ greatly in size, show rates or percentages as well as raw counts, so a large district does not look better or worse simply because it is large.

The UNECE guide says the data source should be identified at the bottom of the chart, and that footnotes can explain definitions or methods.¹ It also recommends using data labels rather than a legend whenever possible, and warns that graphic features that show nothing about the data make a chart less legible.¹ In our own practice, we find that removing gridlines, shadows and decorative icons often does more for clarity than any change of colour.

Many readers see colour differently, use screen readers or read reports on small phone screens. The Web Content Accessibility Guidelines (WCAG) apply to charts as much as to text. Success criterion 1.4.1 says colour must not be the only visual means of conveying information.⁵ In practice, this means labelling lines directly or adding patterns or markers, rather than relying on a colour key alone.

Success criterion 1.4.11 requires parts of graphics needed to understand the content, such as the lines of a graph or the edges of pie slices, to have a contrast ratio of at least 3:1 against adjacent colours.⁶ For charts published online, the W3C images tutorial treats graphs and charts as complex images that need a two-part text alternative: a short description that identifies the chart and a longer description, which can include a data table.⁷

Honest charts are easier when they are planned early. Agree a small set of chart styles for the whole report, with fixed colours for each programme or region, so readers do not have to relearn the key on every page. Ask the person who owns the data to check every chart before design is final, especially axes, units and totals.

Whether you build charts in house or work with a team that offers data visualisation services, ask for the underlying data file with every chart. It makes corrections quick, lets you publish the data table alongside the graphic and gives auditors and partners a clear trail from figure to source.

• Write the chart's message in one sentence before designing it
• Match the chart type to the question: comparison, trend or share
• Start bar and column charts at zero
• Avoid 3D effects and dual axes
• Label values directly and state units and time period
• Put the data source at the bottom of the chart
• Do not rely on colour alone and check 3:1 contrast
• Provide a text description or data table for online charts