Why the Design of a Bar Chart Matters
A bar chart compares values across categories by using the lengths of bars. Because viewers judge those lengths visually, design choices can change how large a difference appears even when the underlying data stay the same. A truncated vertical axis, unequal bar widths, or confusing scale markings can make a modest difference look dramatic, or make the graph difficult to interpret accurately.
The basic rule is straightforward: in a conventional bar chart, bars should start at zero, have equal widths, and be plotted against a clear, evenly spaced numerical scale. These choices make the length of each bar correspond directly to its value. When a chart departs from them, the reader may not be comparing the values in the way the graph seems to invite.
Why Bar Charts Usually Need a Zero Baseline
A bar’s length is a visual measure. If a bar represents a count of 42, a zero baseline gives it a length corresponding to all 42 units. If the axis instead begins at 35, its visible length corresponds only to \(42-35=7\) units. The display may still label the endpoint “42,” but the bar’s length no longer shows the full value.
The effect can be described precisely. Suppose two positive values are \(b_1\) and \(b_2\), with \(b_2>b_1\), and the chart begins at a positive baseline \(L\), where \(0<L<b_1\). The actual ratio of the values is \(b_2/b_1\). The ratio of the displayed bar lengths is instead \((b_2-L)/(b_1-L)\).
To verify the inequality, both denominators are positive, so cross-multiplication preserves its direction:
After expanding both sides and cancelling \(b_1b_2\), this becomes \(-b_1L>-b_2L\). Since \(L>0\), that inequality holds exactly when \(b_2>b_1\). Thus, a positive truncated baseline makes the larger bar look larger relative to the smaller bar than the values actually are.
Worked Example: A Truncated Baseline Exaggerates a Small Difference
Imagine a hypothetical bar chart comparing two categories with counts of 40 and 42. The chart’s vertical axis begins at 35 and shows both bars starting there. Calculate the actual relative difference and compare it with the difference suggested by the visible bar lengths.
Actual comparison. The difference is \(42-40=2\) counts. Relative to the count of 40, that is:
So the count of 42 is 5% greater than the count of 40.
Visible lengths on the truncated chart. The bars extend \(40-35=5\) and \(42-35=7\) units above the displayed baseline. The longer visible bar is therefore:
The visible lengths make the difference appear to be 40%, although the values differ by only 5%. A corrected conventional bar chart starts the axis at zero and shows bars of lengths 40 and 42, with clear labels and a scale that includes both values. The axis should show the full values, not just a narrow range around them.
This does not mean every graph must show every possible value on an axis. The key point is that a conventional bar chart encodes values by bar length, so the zero baseline gives that length a clear reference. If the chart’s purpose is to show small changes within a narrow range, choose a display suited to that purpose and make its scale unmistakable rather than letting truncated bars imply full-length comparisons.
Equal Widths Keep the Visual Encoding Consistent
A conventional bar chart encodes a category’s value using bar height or length, not its area. Keeping widths equal makes that visual convention consistent: twice the value means twice the bar’s height, while the width stays fixed. If widths differ, the bars also occupy different amounts of space from side to side. A wider bar can attract more attention and make its area look disproportionately large.
For a rectangular bar, the visible area is its width multiplied by its height. Therefore, if both width and height change from category to category, the area changes according to both quantities. A reader may not consciously calculate area, but the chart is no longer using one consistent visual size to represent each category’s value.
Worked Example: Unequal Widths Add a Second Visual Signal
Suppose a hypothetical chart compares counts of 24 and 30. The first bar is 1 centimetre wide; the second is 1.5 centimetres wide. Assume both bars start at zero and their heights are proportional to their counts. Compare the count ratio with the ratio of the rectangular areas.
Ratio of counts. The count of 30 is:
That is 25% greater than 24. With equal widths, the bar heights alone would show this 1.25-to-1 ratio.
Ratio of areas. Using the stated widths and heights proportional to the counts, the areas are proportional to \(1\times24=24\) and \(1.5\times30=45\). Their ratio is:
The second bar’s area is 1.875 times the first bar’s area, even though its count is only 1.25 times as large. This display makes width an unintended second visual signal. To revise it, use equal-width bars and a zero baseline: the heights should represent 24 and 30 on the same clear scale.
Scales Must Be Clear and Consistent
A reader should be able to tell what the axis measures, what units it uses, and how the tick marks relate to one another. If equally spaced tick marks represent unequal numerical steps, the axis is not a consistent linear scale. If tick labels omit units or are crowded or ambiguous, the reader may not know what values the bars represent. Both problems weaken the comparison, even if the bar heights were drawn carefully.
Worked Example: Unequally Spaced Values on Equal Tick Intervals
Imagine a hypothetical chart of category counts with vertical-axis labels 0, 10, 20, and 40 placed at equal distances apart. The bars for counts 20 and 40 reach the ticks labeled 20 and 40, respectively. Explain the problem and revise the axis.
The numerical step from 0 to 10 is 10 counts, as is the step from 10 to 20. But the step from 20 to 40 is 20 counts. If all three steps are drawn as equal vertical distances, equal distances on the page no longer represent equal changes in count. A bar at 40 will not be twice as tall as a bar at 20, despite the count of 40 being twice the count of 20.
Revision. Use an evenly spaced scale such as 0, 10, 20, 30, and 40. There are now four equal numerical steps between zero and 40. A count of 20 reaches two steps above zero, and a count of 40 reaches four; the second bar is twice as tall as the first, matching the ratio of their values. Label the axis “Count” and include all tick values so the scale and units are apparent.
A scale does not have to use steps of 10. It could use evenly spaced steps of 5, for example, if that range is useful for the data. What matters is consistency: equal distances between tick marks must correspond to equal numerical increments. Tick labels should also be large enough and clear enough to read, and the axis title should state what is measured. When the graph shows relative frequencies rather than counts, label that scale as a proportion or percentage as appropriate; do not leave the reader to infer the units.
A Practical Check for Bar Charts
When reviewing or revising a bar chart, check each part of the visual encoding rather than relying only on its overall appearance. The following sequence focuses on whether the picture supports a fair comparison of the values.
Check whether the axis shows counts, relative frequencies, or another quantity, and note the units.
For a conventional bar chart, bars should start at zero so their lengths represent their full values.
Make bar widths equal so width does not introduce an unintended visual difference between categories.
Confirm that tick marks are evenly spaced, numerical increments are consistent, and the axis title and units are clear.
Verify that bar lengths, labels, and the underlying values agree, and that the correction has not changed the data.
Common Mistakes and Why They Matter
- Thinking that printed endpoint labels fix a truncated bar chart. A bar labeled 42 still begins at the chart’s baseline. If the baseline is 35, its visible length represents seven units, not 42. Restore the zero baseline for a conventional bar chart.
- Assuming only bar height affects what viewers notice. Unequal widths change the bars’ visible areas and add an inconsistent visual feature. Use equal widths.
- Reading equal tick spacing as proof of an equal numerical scale. Check the labels themselves. Equal distances must represent equal increments; labels 0, 10, 20, and 40 cannot be placed at equal intervals on a linear scale.
- Leaving the scale’s units unstated. A vertical axis without a clear title can make it unclear whether the bars show counts, proportions, or percentages. State what is measured and include its units.
- Correcting the picture but changing the data. A revision should fix the baseline, widths, or scale—not alter category values to make the bars look more balanced.
These checks matter because bar charts are often used to summarize categorical data quickly. A misleading design can shape a reader’s impression before they inspect the labels or numbers. A zero baseline, equal widths, and a clear linear scale do not guarantee that a chart answers every question well, but they make its basic comparisons more faithful to the displayed values.
Check Your Understanding
Use the ideas in this tutorial to assess each display.
- A bar chart shows counts of 50 and 55, but the vertical axis begins at 45. What are the two visible bar lengths, and why might the chart exaggerate the difference?
- Two bars represent counts of 16 and 20. If their widths are 1 centimetre and 2 centimetres, respectively, calculate the ratio of their areas, assuming heights are proportional to the counts. How does that compare with the ratio of the counts?
- An axis has tick labels 0, 5, 10, and 20 at equal vertical intervals. What is wrong with this scale, and how could it be revised?
- Why should a conventional bar chart generally begin at zero, even when the category values are all much larger than zero?
- Name three changes that would make a bar chart with a truncated baseline, unequal widths, and an unlabeled vertical scale easier to interpret fairly.