The core idea
Start with the measure, unit, population and period. Read axis values before judging visual size. Check the underlying numbers and missing observations, then describe only the comparison the chart supports.
1. Read the contract made by the labels
Imagine a chart titled “Our library is growing”. Growing in what: books, registered readers, visits or completed lessons? A useful chart names the measure and unit, the group or place, and the dates. “Weekly visits to a fictional reading room, September, count of visits” is specific enough to begin reading. Visits are not unique people: one reader can visit repeatedly.
Check whether each point covers a day, week or month. A monthly total should not be compared directly with a weekly total as though both measure the same exposure. Also check units: ₹, thousands of rupees and lakhs of rupees can differ by factors of a thousand or a hundred thousand.
2. Bars compare quantities; lines connect ordered observations
A bar chart compares categories by length. Equal-width bars starting at the same baseline make 60 visits visibly twice 30. A line chart places observations in an order, often time, and connects neighbouring points to help show change. A connecting segment does not mean every value between the observations was measured.
Use the question to choose a chart. To compare three reading rooms during the same week, bars are natural. To follow one room across weeks, a line can show a sequence. Neither chart automatically adjusts for opening hours or room capacity. Those differences may matter to a fair interpretation.
Sources: Office for National Statistics: Data visualisation principles ↗
3. Worked example: the cut-off bar
Invented data: 40 visits in week one and 50 in week two. The count rises by 10, or 25%. On a zero-based scale, the bar lengths are 40 and 50 units. If someone starts the axis at 35, the visible lengths become 5 and 15 units. The second bar now looks three times as long even though the second value is only 1.25 times the first.
This is why quantity-encoding bars need a zero baseline. For a line chart, a narrower labelled vertical range can help reveal small changes because position, rather than a filled bar length, encodes the value. Read the actual labels either way. A non-zero axis is not by itself proof that every kind of chart is deceptive.
A 25% increase, shown from zero
| Week | Visits |
|---|---|
| 1 | 40 |
| 2 | 50 |
Sources: Office for National Statistics: Axes and gridlines ↗
4. Check spacing and compare like scales
On a linear scale, equal distances represent equal numerical differences. Labels 0, 10, 20 and 30 at equal distances are linear. Labels 1, 10, 100 and 1,000 at equal distances are a logarithmic pattern: each step multiplies by 10. A log scale can be useful over a wide range, but a straight-looking trend on it is not the same as a constant increase in counts.
Two charts with the same physical height may use different maximum values. A bar almost reaching the top of a 0–50 chart can be smaller than a short bar on a 0–500 chart. Compare values, not screenshot height. In time charts, unequal gaps between observation dates also deserve attention before you infer a rate of change.
Sources: Office for National Statistics: Axes and gridlines ↗
5. An empty observation is not zero
Suppose recorded visits are 40, 50, missing, and 45 for four weeks. Replacing missing with 0 invents an observation. The observed three-week total is 135 and the observed-week mean is 45. You cannot claim that 135 is the full four-week total unless you know there were no visits in the missing week.
A visible gap or a clear missing-data note preserves that uncertainty. Joining points across the gap may suggest a known path that was never observed. If you estimate a missing value, label the estimate and explain the rule; keep the measured and estimated values distinguishable.
Sources: Office for National Statistics: Data visualisation principles ↗
6. Describe a pattern without claiming a cause
For the observed weeks, write: “Visits were 40, 50 and 45 in weeks one, two and four; week three was not recorded.” This is stronger than “our campaign worked”, because the chart alone does not compare what would have happened without the campaign. School holidays, weather or opening hours might have changed too.
A careful caption includes the measure, main numerical comparison and one relevant limitation. Keep the source table so the reader can check the chart. If two quantities move together, that is a starting point for investigation, not the end of a causal argument.
PUT IT INTO PRACTICE
A short investigation with answers
- A bar chart shows values 80 and 100 but starts at 60. Compare values and visible lengths. Answer: values have ratio 1.25; visible lengths 20 and 40 have ratio 2.
- Data are 6, missing and 12 visits. Find the mean of recorded observations. Answer: (6 + 12) ÷ 2 = 9. The mean over all three periods is unknown.
- An axis reads “₹ thousand”; a point is 24. Answer: ₹24,000, not ₹24. Write the unit in your description.
- A campaign starts as visits rise from 40 to 50. Can the chart prove the campaign caused the rise? Answer: no; you need evidence that separates its effect from other changes.
Check your understanding
Why do bars normally start at zero?
Bar length represents quantity; cutting the baseline distorts relative lengths.
Must every line chart start at zero?
No. A clearly labelled narrower range can reveal variation, but the reader must interpret its scale.
Do repeated visits count unique people?
No. One person can create several visits; the measures are different.
Does missing mean nothing happened?
No. It means the value was not available, which differs from a measured zero.
What do equal steps mean on a base-10 log scale?
Each step multiplies the value by 10, rather than adding a fixed amount.
What should a caption avoid?
Unmeasured causes, hidden unit changes and conclusions broader than the displayed evidence.