The graph with a cut axis

A chart showed support for a proposal rising from 48 to 52 per cent. The bars appeared to triple in height, because the vertical axis began at 47.

Starting an axis above zero is sometimes necessary and sometimes deceptive, and the difference is not about honesty but about what the quantity is. For a measurement with a meaningful zero — a count, an amount of money, a number of people — the height of a bar is read as the size of the thing, so cutting the axis breaks the correspondence. For a quantity where zero is irrelevant, such as body temperature or the year in a time series, insisting on zero wastes the whole plot and hides the variation that matters.

The practical rule used by most statistical style guides is narrower than the usual advice. Bars must start at zero, because a bar encodes a length. Lines need not, because a line encodes a slope, and a slope is read as a rate of change rather than a size. The chart in question used bars, which is why it misled.

There is a second detail worth noticing: the four-point change may well be inside the survey's margin of error. A chart cannot show significance, and a reader who corrects the axis and still sees a rise has only corrected half the problem.

The habit is quick. Look at the axis before the shape. If a bar chart's axis does not start at zero, mentally rescale it; if the article does not say what the margin of error is, treat small movements as noise. Both checks take a second and together they remove most chart deception without any statistical training at all.