Traps in Visual Data

The questions on visual data rarely ask you to do math. They ask you to catch a gap between what a chart shows and what the caption claims. Every number printed can be accurate and the chart can still be built to deceive, because deception usually lives in the axis, not the data. Start a vertical axis at 48 instead of 0, and a rise from 48 to 52 percent fills the whole panel, looking like a surge rather than a four-point move. The numbers are true. The scale is lying.

The same trick works through units. Raw counts favor whichever group is bigger, since a larger population almost always produces more cases, more sales, more incidents. Switch to a rate, such as cases per thousand people, and the smaller district can easily come out ahead, because the rate divides out population size and asks a different question: how dense is this, not how big is this. Neither count nor rate is wrong. They answer different questions, and a passage can pick whichever one makes its claim look stronger. Unequal bar widths drawn uniformly cause a related problem, hiding the fact that categories being compared are not the same size to begin with.

The other trap sits between two series plotted on one panel. If broadband subscriptions rise and house prices rise over the same ten years, the figure has shown two separate increases and nothing more. No arrow connects them. The passage supplies the data, your job is to stop exactly where the data stops, and resist a causal story the chart never drew.