Choosing the Right Chart
A decision reference for selecting the chart that fits a message: table versus graph, which chart type matches which comparison, and when to use a map. This page addresses what to choose and why. For building each chart in code, see the Python lessons in this section.
- The message determines the chart, not the reverse. Choose the form that shows what the reader most needs to see.
- Tables and graphs serve different purposes. Tables present exact values; graphs reveal patterns.
- A small set of chart types covers most needs. Bars for comparison, lines for change over time, scatter plots for relationships.
The first decision: Table or Graph
The first question is whether a chart is needed at all. Tables are underused and are often the clearer, more information-dense choice.
| A table is preferable when… | A graph is preferable when… |
|---|---|
| Exact numerical values are required | A pattern or trend is the point |
| Readers will make several localized comparisons | A relationship between variables matters |
| Only a few numbers are involved | The shape of the data is the message |
Presenting a table and a chart together is occasionally the best option, but rarely the default.
Matching the chart to the message
Once a graph is warranted, the type of comparison should determine the type of chart.
| When the message concerns… | Use a… |
|---|---|
| Comparison across categories | Bar chart (horizontal when labels are long) |
| A part of a whole | Bar or stacked bar chart (rarely a pie) |
| Change over time | Line chart |
| A component of a total over time | Stacked area chart |
| A relationship between two variables | Scatter plot |
| A relationship with a third dimension | Bubble chart |
| Geographic variation | Map |
The eye judges length far more accurately than area or angle. A pie chart requires readers to compare slices by area, a task they perform poorly, particularly beyond two or three categories or when three-dimensional effects are added. A bar chart almost always conveys the same shares more accurately.
The three workhorse charts
Most decision-focused communication relies on three chart types. What matters here is recognizing which message each one answers; the code to build them appears in the Python lessons of this section.
- The bar chart answers “which is larger?” It compares values across categories. The axis should start at zero, bars should be sorted by value unless they have a natural order, orientation should turn horizontal when labels are long, and the bar that carries the message should be highlighted.
- The line chart answers “how did it change?” It shows change over time. Points are connected in time order, the number of lines is kept small, and each line is labeled at its end rather than through a legend. Many series are better shown as small multiples.
- The scatter plot answers “are these two related?” It shows the relationship between two variables. A fitted line is added only when it helps the reader see the pattern, and a third variable is encoded through size or color only when it is relevant.
When to use a map
- A map is appropriate when the message is genuinely geographic: when location itself matters, or when the audience needs to locate results within its own territory. Maps are useful for advocacy, since a local official can immediately find their district.
- A map is the wrong choice when geography is incidental, as a map that largely tracks population density can mislead.
Labeling a chart to stand alone
Whatever the chart, three habits let it stand alone:
- The title should state the takeaway (“Deworming added the most school days”).
- Lines and bars should be labeled directly rather than through a legend.
- Axes should carry plain-language names.
A reader who sees only the chart, with no surrounding text, should still grasp the point.
In a phone-based tutoring evaluation in the Philippines, the headline result, that mastery of basic numeracy rose 15 percentage points after only 2.5 hours of instruction, was presented with a single before-and-after bar chart, direct labels, and a title stating the finding. It used no legend, no third dimension, and no pie. The chart type matched the message, a comparison of two values, and that was sufficient for education partners to act.
References
Evergreen, Stephanie D. H. 2019. Effective Data Visualization: The Right Chart for the Right Data. 2nd ed. Thousand Oaks, CA: SAGE.
Schwabish, Jonathan A. 2014. “An Economist’s Guide to Visualizing Data.” Journal of Economic Perspectives 28 (1): 209–34. https://doi.org/10.1257/jep.28.1.209.
Wilke, Claus O. 2019. Fundamentals of Data Visualization. Sebastopol, CA: O’Reilly. https://clauswilke.com/dataviz/.
Additional Resources
Datawrapper Academy. “What Chart Should I Use?” https://academy.datawrapper.de/.