Spot chart choices that exaggerate a change or hide it.
Most misleading charts contain no false numbers. Every value can be accurate, and the person who made the chart may have had no wish to mislead anybody. The problem usually starts with something small: a spreadsheet default nobody changed, a date range cut down to fit one slide, or two data series put on one picture because it looked neat. The people in the room take a message from the shape of the chart that the data doesn't support.
Four design choices cause most of these charts, and each one can be caught with a short question.
On a bar chart, the length of each bar stands for the size of the value, and your eye compares lengths. That only works if every bar starts from zero. An axis that starts at zero is a zero baseline. If the axis begins higher up, the chart shows only the part of each bar above the starting value, and small differences look large.
Priya is an analyst at a mid-sized insurer in Singapore. At a quarterly review, a slide compares customer satisfaction at her company with a competitor. Her company's bar is less than half as tall as the competitor's, and the room goes quiet. Priya checks the left edge of the chart. The axis begins at 80, and the scores are 82 and 85 (example figures).
A score of 85 is only about 4 percent more than 82. With the axis at 80, Priya's company gets a visible bar of 2 units and the competitor gets 5 units, so one bar looks two and a half times the other although the gap is three points.
Priya doesn't accuse anyone. She says the axis begins at 80, so the comparison is 82 against 85, and asks whether that gap is bigger than the usual movement from one quarter to the next. The room relaxes, and the discussion turns to whether three points is a real difference or ordinary noise. You can use the same follow-up whenever you notice a scale problem, because it moves the meeting from the picture to the numbers.
The reverse trick works too. If the axis runs far above the data, say from 0 to 500 for scores in the 80s, real differences shrink to nothing.
Line charts follow a different rule. A line shows change over time, and forcing the axis to start at zero can flatten a meaningful movement into a straight line. Share prices and monthly error rates are often plotted with the axis starting near the data, and that can be fine. What a line chart needs is a clearly labelled axis, so you can tell how much change a given slope represents. For bars the rule is firm: start at zero, or don't use bars.
Before you react to any chart, read the axis labels. It takes two seconds and catches most of the problems described here.
A dual-axis chart puts two lines on the same picture, with a different vertical axis on each side. A common version plots sales on the left axis and marketing spend on the right. When the lines rise and fall together, the chart implies that spend drives sales.
Whoever built the chart chose both scales. By stretching or squeezing either axis, you can make almost any two rising lines look as if they move in step. The overlap is partly a design decision, and a dual-axis chart is no evidence that one thing drives the other. Module 4, on correlation and causation, covers whether one thing really drives another.
When you see one of these charts, ask for each line on its own chart with its own axis, or ask for the actual numbers.
Every chart over time has a first and last point that someone picked. If you start at a low point, almost anything looks like a rise. Start at a peak and it looks like a fall.
Seasonal patterns can pass for growth. Picture a monthly sales chart running from March, just after the slow Chinese New Year period, to November. The line climbs steadily. Extend it to three full years and you may see the same shape every year: a dip after the new year and a climb towards December. In that case the apparent growth came only from the calendar.
Two questions catch this: why does the chart start here, and what would it look like over a longer window? If earlier data doesn't exist, that's fine. If it exists and was left off, the chart tells you less than it seems to.
Totals grow with population, so when you compare totals across groups of different sizes, you may only learn which group is bigger. Typical cases are burglaries in one town against another, or customer complaints by company region.
Town A records 300 cases in a year and has a population of 150,000. Town B records 400 cases and has 400,000 people. If you put the totals on a bar chart, Town B looks worse.
A per person figure is a count divided by population, here expressed per 100,000 people. For Town A, 300 / 150,000 x 100,000 gives 200 cases per 100,000 people. For Town B, 400 / 400,000 x 100,000 gives 100 cases per 100,000 people. Town A's rate is twice Town B's, so the per person comparison reverses what the totals chart showed.
Whenever totals are compared across groups of different sizes, or across years when the population changed, ask whether a per person figure would tell a different story. The same goes for anything that grows with size: total sales across branches of different sizes, total complaints across products with different numbers of users, and total road accidents as more cars come onto the roads.
Before you respond to what a chart seems to say, run these four checks:
Check the axis: read the labels first, and make sure bars start at zero. Check the window: ask why the chart starts and ends where it does, and what a longer window would show. Ask whether it should be per person: convert totals to rates when groups or populations differ in size. Be careful with two axes: ask for each line on its own chart with its own axis, or for the actual numbers.
Ask calmly and specifically, as Priya did, without accusing anyone. After the four checks, the discussion can move to what the numbers say, such as whether a gap is a real difference or ordinary noise.
The fastest way to learn this is to redraw a chart yourself and see how different it looks. In the activity you will find one chart in the news or a company report, note which of the four checks it fails, and rebuild it in a spreadsheet with that choice fixed.
Find a chart in a news article or company report and note which of the four checks it fails. Redraw it in a spreadsheet with that choice fixed, and write one line on whether the original exaggerated or hid the change.
Junxiong-WFG Organisation is an authorised representative of AIA Financial Advisers Private Limited (Reg. No. 201715016G).