The same dataset supports several very different-looking charts, and readers draw different conclusions from each.
Which makes visualisation choices analytical decisions rather than aesthetic ones — and they should be made with that in mind.
Choices that change interpretation
- Axis range — a truncated axis exaggerates differences dramatically.
- Whether zero is included, which matters for bar charts especially.
- Which subset is plotted.
- Whether uncertainty is shown.
- Chart type.
- Colour scale for continuous data.
Common distortions
- Bar charts with truncated axes.
- Showing means without distributions, hiding overlap entirely.
- Omitting error bars or confidence intervals.
- Inconsistent scales across comparable panels.
- Selecting the time window where the pattern is strongest.
- Most of these are unintentional and all are visible to reviewers.
Honest defaults
- Start bar chart axes at zero.
- Show all data points where the sample is small enough.
- Show distributions rather than only summary statistics.
- Show uncertainty and say what kind it is.
- Use the same scale across panels being compared.
- Plot the full range unless there is a stated reason not to.
Accessibility
- Do not rely on colour alone — use shape or pattern too.
- Check it works in greyscale, since figures get printed.
- Ensure sufficient contrast.
- Check legibility at printed size, not on screen.
Testing your own
- Show it to someone with the caption and no text — can they say what it shows?
- Ask whether a different reasonable choice would tell a different story.
- If yes, say why you chose this one.
- Have a colleague check for unintentional distortion.
In aggregate reporting
- State the counting rule with any chart of aggregates.
- Show coverage limitations.
- Do not present a trend that spans a methodology change without flagging it.
- Distributions are more honest than single figures.
One thing worth remembering
Ask whether a different reasonable presentation choice would tell a different story.
If it would, you have made an analytical decision — and readers are entitled to know it was made, and why.
Câu hỏi thường gặp
Which choices change interpretation?
Axis range and whether zero is included, which subset is plotted, whether uncertainty is shown, chart type, and colour scale for continuous data.
What are the common distortions?
Truncated bar chart axes, showing means without distributions, omitting error bars, inconsistent scales across comparable panels, and selecting the strongest time window.
What are honest defaults?
Bar chart axes starting at zero, showing all data points where possible, showing distributions rather than only summaries, showing uncertainty with its type, and consistent scales.
What accessibility checks are needed?
Not relying on colour alone, checking it works in greyscale since figures get printed, ensuring sufficient contrast, and checking legibility at printed size.
Which question tests a chart?
Would a different reasonable presentation choice tell a different story — if so, you have made an analytical decision readers are entitled to know about.
