---
name: Truncated axis
slug: truncated-axis
category: pattern
status: published
created: 2026-08-21T00:00:00.000Z
modified: 2026-08-21T00:00:00.000Z
definition: Starting a value axis above zero so differences look bigger than
  they are; honest for lines and dots, a lie for bars.
aliases:
  - name: broken axis
  - name: axis truncation
  - name: cut y-axis
  - name: non-zero baseline
  - name: zero baseline
tags:
  - dataviz
  - perception
relations:
  contrastWith:
    - axis
    - dual-axis
  variantOf: []
  partOf: []
  seeAlso: []
implementations: []
sources:
  - title: "Calling Bullshit: Misleading axes"
    url: https://www.callingbullshit.org/tools/tools_misleading_axes.html
  - title: "Fundamentals of Data Visualization: Proportional ink"
    url: https://clauswilke.com/dataviz/proportional-ink.html
demo: inline
exhibit: false
useWhen: judging whether cutting the y-axis exaggerates or clarifies
---

A value axis does not have to start at zero, and most of the long-running argument about
misleading charts is really an argument about when it does. Truncating means beginning the
scale just under the smallest number in the data, which spends the whole plot on the range
those numbers actually occupy. Every difference gets bigger. Whether that reads as
clarification or as exaggeration depends entirely on what the marks encode, which is why
"always start at zero" is a rule that is right half the time and confidently wrong the
other half.

Bars encode value as length measured from a baseline, so the baseline is part of the
encoding rather than a framing choice around it. Claus Wilke states the constraint as
[proportional ink](https://clauswilke.com/dataviz/proportional-ink.html): the ink a mark
spends has to be proportional to the value it stands for, and a bar cut off underneath is
spending its length on a quantity nothing in the chart names. The specimen above runs the
same four quarters both ways. Handset share moves from 46.8 to 48.4 percent, a rise of 3.4
percent, and with the axis cut at 46.4 the last bar is drawn five times the height of the
first. Nothing in the drawing is false. Every bar ends at the correct tick, and the tick is
labelled. The reader is simply being invited to compare lengths that have stopped meaning
what lengths mean.

Lines and dots are the other case, and they are the reason the zero rule cannot be applied
blind. A line encodes value as position, read against the labelled scale rather than against
the bottom of the frame, so a chart of body temperature, of a share price, or of pH has no
business starting at zero: the plot would be a flat band with every reading crushed into a
few pixels. Truncation there is not a trick, it is what makes the chart legible at all. The
dishonesty is never the cut. It is cutting silently, with a scale small enough to go
unread and a shape loud enough to be remembered, which is what the axis break glyph (the
little zigzag through the baseline) exists to prevent. [Calling
Bullshit](https://www.callingbullshit.org/tools/tools_misleading_axes.html) collects the
cases where that signal was left off on purpose.

Three neighbours separate this cleanly. [Axis](/axis) is the component a truncation is a
choice about: the axis is the drawn line, its ticks, and its labels, while truncation is a
decision about the domain those labels report, so the same axis component draws both charts
in the specimen. [Chart](/chart) is the whole drawing the axis serves. And when the cut is
made deliberately, to win an argument the numbers do not support, the word for it stops
being a charting term: that is a [dark pattern](/dark-pattern) working by
[misdirection](/misdirection), sending attention to a shape while the evidence sits in
small grey type at the left edge.