Starting a bar chart's axis above zero turns a tiny real difference into a dramatic one

What it is

Starting a bar chart's y-axis at a nonzero value (beginning at 90 to display data between 90 and 105) turns visually tiny numeric differences into dramatic height differences: two groups actually differing by 5% look three times apart on a truncated axis. Truncation changes no numbers, but it changes the visual presentation of "difference"—the information readers actually care about—and the overwhelming majority never check where the axis starts. It is the classic, easiest-to-execute misleading technique and a standard case in misleading-visualization research.

Why it happens

The exaggeration mechanism is the conflict between length encoding's perceptual properties and the axis's implicit promise. Readers decode bar charts by height difference (the length channel), and the perceptual system processes "height ratio = value ratio"; a truncated axis breaks that equality—on an axis starting at 90, the bar at 100 is half the height of the bar at 110, while numerically 100 is only 9% smaller. The interpretation guided by perception is disconnected from the true numeric ratio, and the axis tick labels are too small and inconspicuous to correct the first impression. The exaggeration grows with the truncation depth: the farther the data range sits from the axis origin, the larger the visual amplification (an axis from 90 can magnify a 5% difference into a visual ratio over tenfold). A key distinction: truncation's acceptability differs between line charts (position encoding) and bar charts (length encoding)—lines are read by relative position and direction, so a nonzero origin is common and acceptable (extending into the same group's baseline rule), while bars and areas encode by length, for which a zero baseline is the mathematical precondition, and truncation produces proportional distortion.

Where it stops holding

Truncation is not always misleading: when all data sits in a narrow band far from zero and between-subject differences are the point (daily stock movement, body temperature changes), a truncated axis is the necessary means of making differences legible—the criterion is whether the graphic will be "read as proportions." Compliant practice has three elements: explicitly mark the truncation (axis-break symbol or a labeled origin), avoid length encoding (use points or lines instead of bars), and state conclusions numerically rather than by visual impression. A full view of the same data (from zero) can be offered via interaction or an inset so first impressions coexist with complete context. Truncation's harm also scales with purpose: in internal exploration (looking at your own data) it is a tooling choice; in outward communication (persuading others) it slides from tooling into rhetorical manipulation.

Applying it

• Default bar and area charts to zero origins; when narrow-range detail must show, switch to dot or line plots and explicitly label the axis origin.
• Mark any nonzero origin on the axis itself (a wave break symbol or "axis starts at nonzero" note).
• Run a dual check before publishing: do bar-height ratios match numeric ratios? A mismatch means proportional distortion.
• Verification: ask readers to estimate the groups' difference multiple from the graphic alone; an answer off by more than one multiple from the true value means the chart misleads.

Related

• Same group: U10.01.2 Stretching the aspect ratio changes the impression of trends · U10.01.3 A nonzero baseline is an error under length encoding
• Nearby: U4.03.1 Truncating the y-axis exaggerates differences · U2.02.2 Baselines must start at zero
• Search terms: truncated axis · misleading chart · axis manipulation