Team A scores 48 points and Team B scores 50. Drawn from 0, the bars are 48 and 50 units tall, almost the same. Now start the vertical axis at 47 and stop it at 51. Team A's bar shows only 48−47=1 unit and Team B's shows 50−47=3, so B's bar looks three times as tall. The real difference is 2 points, about 4%. Zooming in on the tops of the bars blew it up.
The trick also works the other way. Plot a town's population rising from 10,000 to 12,000 on an axis that runs from 0 to 1,000,000, and the line looks flat, hiding 20% growth. Uneven intervals are the second kind of scale distortion: an axis labelled 0, 10, 20, 50, 100 at equal spacing squashes the large values together, so a fast-growing trend can look as if it is slowing down.
Not every axis that skips zero is a trick. On a bar graph the bar's length stands for the amount, so it should start at zero. On a line graph the shape shows change, and starting near the data is often the honest choice: a fever chart from 36 to 40 °C shows a dangerous rise that a 0 to 40 °C axis would hide. Scale distortion is also only one kind of misleading graph; leaving out bad months is cherry-picking, a different trick, even on a perfectly drawn axis.
Think of it like this. Scale distortion is like the zoom on a camera. Zoom in on a small bump in the road and it fills the photo like a hill; zoom far out from a mountain and it shrinks to a pebble on the horizon. The bump and the mountain did not change, only how close the camera got. The axis numbers tell you how far the graph has zoomed.
Core idea. Scale distortion is changing where an axis starts or how its intervals run to fake the size of a difference.