Five ways a chart can mislead — and the honest version of each.
Every fix below is tied to a named visualization principle, organized by Munzner's three questions: why the chart exists, what data it shows, and how it encodes that data. Each chart is our own reconstruction on neutral, illustrative numbers — the techniques, not anyone's actuals.
Does the chart type serve the question being asked?
When the task is ranking, the idiom has to make rank readable. Some chart types simply can't.
Share of sales by product
Share of sales by product
A pie can't rank near-equal slices. The question is "which product sells most," but the eye judges angle poorly, so the order is unreadable. Sort the values into bars on a common scale and the ranking lands at a glance. Based on the classic finding that position and length beat angle and area for quantitative comparison.
Is the chart showing all the data that matters?
Not every distortion is in the axis — some are in what gets left out. The encoding can be flawless while the selection of data does the misleading.
Performance index, recent months
Performance index, full history
The chart showed only the three months that rose. Cropping the window hides a year of decline — the metric is still below where it started. Plot the full series so the recent uptick reads in context, not as a clean climb. Based on the everyday move of zooming a chart to the window that flatters the story.
Does the visual mapping match the numbers?
The axis, the second axis, the color ramp — the encoding has to let the reader recover the real values. When it doesn't, the chart lies even with honest data.
Quarterly revenue ($M)
Quarterly revenue ($M)
The y-axis started at 90, not 0. On a bar chart, length is the value — so a clipped baseline turns a 6-point rise into a towering one. Start bars at zero and the real story shows: steady, modest growth. Based on the most-taught distortion in visualization: a non-zero baseline on a bar chart.
Spend vs. signups, by month
Signups against spend
Two independent y-axes can manufacture any correlation. Scaling each line to fill the frame forces them into lockstep and hints at a cause the data can't carry. Put the relationship on one honest scale — a scatter — and the real, weak association is visible. Based on the spurious-correlation genre: unrelated series made to look linked by axis choice.
Metric by region and month (illustrative)
Metric by region and month (illustrative)
A rainbow palette isn't ordered by brightness. Equal data steps map to unequal jumps in color, so the eye invents a peak at yellow and a false band at the green edge — and roughly one in twelve readers can't separate the red from the green. A single-hue ramp that darkens with value lets anyone read high from low at a glance. Based on the well-documented retirement of rainbow palettes in favor of perceptually uniform, monotonic ramps.
Be first to put your charts on trial.
FairCharts reviews any chart, flags what's misleading, and hands back a corrected version — with the reasoning to defend it in the room.
Join the waitlist