Research Guides
Check the Bar Chart Baseline Before Writing the Headline
A bar that looks twice as long can tempt an editor to describe a dramatic advantage. Before writing that headline, inspect where the numerical axis begins. The visible length may represent only the portion above a cropped baseline, not the full value.
This is a focused review method for comparison graphics, not a certification of a dataset or a claim about Colquhoun.shop chart controls.
1. Recover the actual values and units
Record the category names, values, units and period from the underlying data, not by estimating bar lengths. Check whether the chart shows counts, rates, money or a derived difference. A bar labeled “change” needs the reference period or baseline that defines that change.
Keep unknown values distinct from zero. If one category has no observation, an empty space should not silently become a measured absence. Confirm that the comparison uses compatible definitions before evaluating the graphic.
2. Inspect the bar origin
The UK's Office for National Statistics advises that a bar chart's axis start at zero. If showing small differences requires an axis above zero, it suggests considering another chart type, such as a dot plot. ONS: Bar chart guidance
Consider invented values of 90 and 100. Starting the displayed bars at 80 produces visible lengths of 10 and 20. The second visible segment is twice the first, but the actual value of 100 is not twice 90. This is why a correct data label alone does not make the bar-length comparison faithful.
3. Choose the graphic to match the question
If the question is about total magnitude, retain the full bar lengths from zero. If the editorial question is specifically about a small difference, consider a clearly labeled alternative that emphasizes positions rather than full bar lengths, or show the difference directly with an explicit reference.
Do not change the underlying metric merely to make the picture more dramatic. A chart of percentage-point changes is not a chart of the original percentages. The caption and headline should name the quantity actually plotted.
4. Check the surrounding presentation
For several side-by-side panels, inspect whether scales are shared or different. Two bars with the same on-screen length can represent different values when the panels use different ranges. Make any intentional scale difference unmistakable and reconsider whether direct visual comparison is appropriate.
Check the exported image as well as the interactive view. Cropping can remove a unit label, axis origin or qualification. A social preview or slide should not tell a stronger story than the complete chart.
5. Save a chart-to-claim record
- Data: categories, values, units, period and source.
- Encoding: what bar length represents and where it starts.
- Scale: range and consistency across panels.
- Claim: the proposed headline and its numerical support.
- Handoff: exported image and labels inspected.
- Outcome: retained chart, revised graphic or unresolved evidence gap.
Pair this with the guide to making a research chart readable without a color key. A faithful baseline does not by itself make a chart accessible or its data reliable, but it removes one avoidable way for a sound number to become a misleading picture.