The reason a graph without zero can double an effect

A bar chart shows quarterly revenue: 98, 100, 102, 104 million. When the y-axis starts at zero the bars look nearly identical. When the axis starts at 96 the bars show a dramatic staircase. The data has not changed; the visual encoding has, and the visual encoding is what the viewer processes first and remembers longest.

What everyone sees

The truncated axis looks like a reasonable design choice: starting at 96 uses the space efficiently and shows the trend more clearly. The viewer may not notice the truncation at all, because the axis label is small and the visual pattern is loud. Even viewers who notice the starting point rarely recalculate the slope in their heads, because the visual impression has already formed.

What is actually happening

Pandey and colleagues experimentally confirmed that truncated axes cause viewers to overestimate the magnitude of differences, even when axis labels are clearly visible. Cleveland and McGill’s foundational work on graphical perception showed that humans decode differences in bar height as ratios, not as absolute lengths. When the baseline is moved from zero to 96, the visible height difference between 98 and 104 is tripled, and the perceived effect size scales with the visible ratio rather than the actual ratio.

Why it stays hidden

The distortion hides because the numbers on the axis are accurate and the data points are correctly plotted. No data has been fabricated, so the graph passes every fact check. The deception lies in the encoding, not the data, and most media-literacy training focuses on the data rather than the encoding. A viewer trained to check whether the numbers are correct will find that they are, and conclude that the graph is honest.