Misleading axes
Same two numbers. Pick where the axis starts and you can make the difference look like anything.
Try it
Drag the baseline of the chart up and down.
Made-up numbers for illustration: Brand A scores 50, Brand B scores 52 (a 4% difference).
What is going on
Bars encode value by length, so the baseline matters: a bar that starts at zero has a length proportional to the quantity. Start at 40 instead and a bar for 55 is three times as long as one for 45, though the quantity is only 22% bigger.
Lines are more forgiving, since they show change rather than size, but a stretched vertical scale can still turn a wobble into a cliff. Whenever bars are used, ask where zero went.
How to spot it
- Axis labels start somewhere other than zero on a bar chart.
- No axis labels at all.
- A chart where the visual change feels much bigger than the number change quoted.
The gee-whiz graph
Darrell Huff devoted a chapter of How to Lie with Statistics (1954) to the “gee-whiz graph”, in which cutting off the lower part of the scale makes small changes look dramatic. Edward Tufte’s “lie factor”, the ratio of the size of an effect shown in a graphic to the size in the data, puts a number on the same distortion.
The two bars in the demo are made-up numbers chosen for illustration.