None of the mistakes in this guide look like mistakes. A truncated y-axis isn't a lie - it's a legitimate zoom-in that happens to make a small change look dramatic. A cherry-picked timeframe isn't fabricated data - it's a real window that happens to be the one where the trend looks best. Each individual choice is small and defensible in isolation, which is exactly why they're so common and so rarely caught in the moment. The damage isn't any single instance - it's what happens the first time a sharp reader checks the underlying numbers and finds a gap between what the chart implied and what the data actually shows. That's the moment a dashboard stops being trusted, and trust, once spent, is expensive to rebuild.
Table of Contents¶
- The Truncated Axis
- The Cherry-Picked Timeframe
- The Pie Chart With Too Many Slices
- Cumulative Charts That Only Ever Go Up
- Dual Axes That Imply a Relationship
- Rounding That Changes the Story
- A Worked Example
- FAQ
The Truncated Axis¶
Starting a y-axis at 90 instead of 0 turns a genuine 2-point move from 94 to 96 into a bar that looks twice as tall - technically accurate, visually dishonest. The underlying numbers on the axis labels are correct; what's misleading is the visual impression, since a reader's eye interprets bar height and line steepness proportionally, and a truncated axis breaks that proportion without saying so. This isn't always wrong - a truncated axis can be the right call for a metric that never moves outside a narrow band (a 98-100% uptime chart, say, where a full 0-100 axis would flatten every real change into invisibility) - but it needs to be a deliberate, disclosed choice, not a default. Labeling the axis start clearly, or adding a short note explaining why the scale is zoomed, is the difference between a legitimate design decision and a quiet distortion.
The Cherry-Picked Timeframe¶
A trend line covering the last six weeks can tell a completely different story than the same metric over the last twelve months, and choosing the window that happens to support the point you're making - even unconsciously, even with good intentions - is one of the easiest ways to mislead without technically lying. This is especially easy to fall into when a dashboard is built to support a specific recommendation: the natural instinct is to reach for the timeframe that makes the case most cleanly, and that instinct is worth catching deliberately rather than trusting by default. A simple habit helps: default to a standard, consistent timeframe (the last four quarters, the last full year) for any tracked metric, and treat any deviation from that default as something that needs an explicit, stated reason rather than a silent choice.
The Pie Chart With Too Many Slices¶
A pie chart works reasonably well for two to four categories, where the relative size of each slice is still easy to judge by eye. Past six or seven slices, distinguishing a 9% slice from an 11% slice becomes genuinely difficult - human perception is much better at judging the length of a bar than the angle of a wedge, which is why a long list of categories almost always reads more clearly as a sorted horizontal bar chart than as a pie. A pie chart with ten-plus thin slivers, several of them visually indistinguishable from each other, isn't dishonest exactly, but it does obscure real differences the underlying data actually contains - a reader walks away with "lots of small categories" as the takeaway, when the real story might be that three of those categories are meaningfully larger than the rest.
Cumulative Charts That Only Ever Go Up¶
A cumulative total - total signups over time, total responses collected - always trends upward by definition, even during a period where new activity has completely stalled. Presented on its own, a cumulative chart can visually imply healthy, continuous growth during a stretch where the underlying rate has actually flattened or declined, since the line keeps climbing even if it's climbing by smaller and smaller increments each period. Pairing a cumulative total with the period-over-period figure it's built from - new signups this month, not just the running total - gives a reader the ability to see deceleration that the cumulative view alone hides entirely.
Dual Axes That Imply a Relationship¶
Plotting two different metrics on the same chart with two separate y-axes - satisfaction on the left, feature usage on the right - visually implies the two lines are related, simply by virtue of sharing a chart. The two axes are independently scaled, though, which means the relative steepness of each line can be manipulated almost arbitrarily just by choosing where each axis starts and ends, making two genuinely unrelated metrics appear to move together, or two genuinely related metrics appear unrelated, depending entirely on axis choices that carry no statistical meaning at all. If two metrics are worth comparing directly, a scatter plot - which actually tests whether they move together, rather than just visually implying it - is a more honest choice; our guide on correlation and causation in survey data covers the deeper reasoning behind why an apparent visual relationship isn't evidence of a real one.
Rounding That Changes the Story¶
Rounding 49.6% up to 50% is defensible in isolation and can become a real distortion once several rounded figures are added, compared, or used to justify a threshold - "we crossed 50%" reads very differently from "we're at 49.6% and rising," even though the second is the more accurate, and often more genuinely encouraging, statement. This matters most around meaningful thresholds - a pass/fail cutoff, a target the organization has publicly committed to - where rounding can quietly convert "we didn't quite make it" into "we made it." Showing one decimal place for any number sitting within a percentage point of a stated threshold is a small habit that closes this gap without cluttering every other number on the dashboard with unnecessary precision.
A Worked Example¶
A customer success team presents a quarterly dashboard showing a satisfaction score climbing from 91 to 96 on a chart with a y-axis starting at 85 - visually, a dramatic, reassuring jump. A board member, skeptical of how steep the line looks for a 5-point move, asks to see the same data on a full 0-100 scale, where the same numbers plot as a nearly flat line. The underlying data was never wrong, but the chosen scale had turned an ordinary, modest improvement into something that looked like a major turnaround, and the mismatch between the visual impression and the real magnitude cost the team real credibility in the room - not because anyone had lied, but because the axis choice hadn't been disclosed or justified, and the gap became the story instead of the actual result.
FAQ¶
Is a truncated axis always wrong?
No - it's appropriate for metrics that genuinely live in a narrow range, where a full-scale axis would flatten every real change into invisibility. The issue is doing it silently rather than disclosing the choice and the reason behind it.
How many categories is too many for a pie chart?
Past six or seven, a sorted horizontal bar chart almost always communicates the same data more clearly, since bar length is easier to judge accurately by eye than the angle of a pie slice.
Should I ever use a cumulative total chart?
Yes, when the running total itself is the thing that matters (total responses collected toward a target, for instance) - just pair it with the period-over-period rate alongside it, so a reader can see whether the underlying pace is accelerating, steady, or slowing.
What's the fix for dual-axis charts?
Where possible, use a single shared axis, index both metrics to a common starting point (both set to 100 at the start of the period), or use a scatter plot if the actual question is whether the two metrics move together.
For more on reading and presenting data honestly, see Is My Survey Result Real, or Just Noise? and Writing Survey Reports People Act On.