You've seen the pattern. Someone spends three days building a dashboard with fourteen metrics, a dual axis, a colour gradient, and a 3D donut. The meeting ends with "interesting, let's circle back." The next week, a colleague shows one bar chart with two bars and a label. The budget gets approved on the spot.
This isn't an accident, and it isn't about taste. There's solid perceptual science, a few famous historical lessons, and some hard-won business experience behind a simple claim: the chart that removes the most gets believed the most.
This post walks through why that's true, when it isn't, and how to simplify a chart without dumbing it down.
How people actually read charts
In 1984, statisticians William Cleveland and Robert McGill ran a series of experiments asking a simple question: which visual encodings do humans decode accurately? They identified ten elementary perceptual tasks (position along a common scale, length, angle, area, volume, colour saturation, and so on), and tested how well people estimated values using each.
The results produced a rough hierarchy of accuracy, from best to worst:
- Position along a common scale (dot plots, simple bar charts)
- Position along non-aligned scales
- Length, direction, angle
- Area
- Volume, curvature
- Shading and colour saturation
The practical implication is blunt. In a bar chart, you only have to look at where each bar ends against a single axis, a comparison your visual system makes in a split second. In a pie chart, you're asked to compare angles and areas, which almost everyone struggles to do accurately. Later replications by Heer and Bostock using online participants produced very similar results.
Every time you choose a "richer" encoding (a bubble instead of a bar, a radial chart instead of a line) you are, in effect, asking your reader to do a harder perceptual task. Harder tasks mean more errors, slower reading, and less confidence. A reader who isn't confident they've read the chart correctly will not be confident in your conclusion.
Five ways complexity weakens an argument
It splits attention. An argument is a single claim. A chart with eight series makes eight claims at once, and the reader has to guess which one you mean.
It invites the wrong question. The moment a chart has a secondary axis, someone will ask about the secondary axis. You wanted to discuss revenue; now you're defending a scale choice.
It hides the gap. Arguments live in differences: before vs. after, us vs. them, expected vs. actual. Decoration, gridlines, and extra categories shrink the visual size of that gap.
It signals uncertainty. A dense chart reads as "I'm not sure what matters, so here's everything." A sparse chart reads as "I know exactly what matters."
It raises the cost of agreeing. Agreement is cheap when the reader can verify the claim in two seconds. It's expensive when they need a legend, a footnote, and a second look.
Edward Tufte gave this a name in 1983. Chartjunk refers to all visual elements in charts that aren't necessary to comprehend the information, or that distract the viewer from it. His companion idea, the data-ink ratio, is the proportion of a graphic's ink that actually carries data. The higher the ratio, the harder it is for a reader to look at anything other than your point.
Write the title as the verdict, not the topic
Most chart titles describe the data: "Quarterly revenue by region, 2024 to 2026." That's a label, not an argument. The reader is left to work out what you want them to notice, and if they notice the wrong thing, your chart has just made someone else's point.
A verdict title states the conclusion in plain language: "Revenue in the East region has fallen for five straight quarters." Now the chart's only job is to let the reader verify a sentence they've already read. Verification is fast; interpretation is slow.
This matters more than it sounds. In meetings, people glance at a slide for two or three seconds before they listen to you. If the title already carries the claim, that glance does your persuading. If the title is a label, the glance produces a question, and you spend your speaking time answering it.
Three rules for verdict titles:
- Use a full sentence with a verb. "Churn doubled after the price change," not "Churn and pricing."
- Put the number in it when the number is the point. "Support load per engineer is up 2.4x" beats "Support load is rising."
- Keep the descriptive label as a subtitle. Date range, units, and source still need to appear; they just move down one line.
If you can't write a verdict title for a chart, that's a diagnostic result: the chart doesn't yet contain an argument.
Case study: The Challenger launch decision
On the night of 27 January 1986, engineers at Morton Thiokol tried to convince NASA to delay the Challenger launch. They believed cold temperatures would stiffen the rubber O-rings that sealed the booster joints. The forecast for the next morning was far colder than any previous launch.
They had the right conclusion and the right data. What they didn't have was a chart that made the argument.
The evidence was presented across thirteen charts, including a handwritten chart indicating concern about the temperature effect on seal actuation time. Launch temperature and O-ring damage were never plotted together on one axis. Managers looked at the material and concluded the temperature effects on the O-rings were inconclusive.
When the Rogers Commission later drew the same data as a single scatter plot (damage on one axis, temperature on the other) the relationship was obvious: every launch below about 65°F showed damage, and the forecast launch temperature sat far to the left of anything ever flown. Tufte, in Visual Explanations, redraws it further and adds one marker for the morning's forecast. The argument needs no narration.
You can read the original charts and testimony in NASA's archive of the Rogers Commission report. It's sobering: the data was sufficient, the people were right, and the presentation failed.
Lesson: the most important variable goes on an axis. If your argument is "X causes Y," plot X against Y. Nothing else.
Case study: Nightingale's rose vs. a plain bar
Florence Nightingale's 1858 "Diagram of the causes of mortality in the army in the East" is one of the most celebrated charts in history. Its radial wedges show, month by month, that far more British soldiers in the Crimean War died from preventable disease than from wounds.
It worked. It helped push through sanitary reforms in military hospitals. But it's worth asking why it worked, because it wasn't the elegance of the wedges. Radial area is near the bottom of the Cleveland and McGill hierarchy; readers consistently misjudge it.
The chart persuaded because the gap was enormous. Disease deaths dwarfed battle deaths by such a margin that even a hard-to-read encoding couldn't hide it. Redraw the same data as two bars ("died of disease" and "died of wounds") and the message hits faster and more accurately.
Lesson: a beautiful chart can carry an argument when the effect is huge. When the effect is modest (which is most of the time), only a simple chart will.
Case study: The dashboard that lost the room
This one is a composite drawn from common consulting experience rather than a single named company, but you'll recognise it.
A product team wants more engineering headcount. Their first attempt is a dashboard: monthly active users, retention cohorts, feature adoption, support ticket volume, and NPS. Five panels, three colour palettes, one dual axis. Leadership's response is a list of clarifying questions about the retention cohorts. No decision.
The second attempt is one line chart: support tickets per engineer over eighteen months, climbing steadily, with a dotted horizontal line marking the level at which SLAs started slipping. One title: "Each engineer now handles 2.4x the tickets they did last year."
Headcount approved.
Nothing changed in the underlying data. What changed was that the chart finally made one claim, put the key ratio on the axis, and annotated the threshold that mattered to the decision-makers. The dashboard was a report. The line chart was an argument.
Lesson: dashboards are for monitoring. Arguments need one chart, one claim, one annotation.
A worked redesign: from twelve series to one claim
Here's how a simplification actually proceeds, step by step. Suppose you start with a monthly line chart of website traffic for twelve marketing channels over two years. It's accurate, complete, and unreadable: a tangle of coloured lines with a twelve-item legend.
Step 1: Name the claim. After a minute of staring, you realise what you actually want to say: "Organic search now brings in more traffic than paid search, and the gap is widening." Write that down.
Step 2: Keep only what the claim needs. The claim mentions two channels. Remove the other ten. The chart drops from twelve lines to two.
Step 3: Put the comparison on a common scale. Both lines already share a y-axis; good. If the claim had been about shares rather than volumes, you'd switch to percentage of total traffic so the comparison doesn't get distorted by overall growth.
Step 4: Mark the moment that matters. The lines cross in a specific month. Add a thin vertical line and a label: "Organic overtakes paid, March 2025." That's the hinge of the argument, so it gets the only annotation.
Step 5: Remove everything that isn't data or the annotation. Gridlines lightened or gone. Legend deleted; label the two lines directly at their right-hand ends. No border, no background, no drop shadow.
Step 6: Write the verdict title. "Organic search overtook paid search in early 2025 and the gap keeps growing." Subtitle: "Monthly sessions, Jan 2024 to Sep 2026. Source: analytics export."
Step 7: Keep the original as a backup. The twelve-line chart goes in an appendix slide. When someone asks "what about email?", you have the answer in one click, and the fact that you had it ready builds trust.
The finished chart has two lines, one annotation, and one sentence. It contains roughly a sixth of the original data and makes perhaps ten times the impact.
A checklist for simplifying without losing rigour
Simplicity is not the same as omission. Use this before you present:
- Write the claim as a sentence first. If you can't, you don't have an argument yet; you have exploratory data.
- Put the claim's variables on the axes. Everything else is context, and context goes in a footnote or a backup slide.
- Choose the highest-accuracy encoding that fits. Bars and dot plots for comparisons, lines for change over time. Reach for angle or area only when you've ruled out position and length.
- Cut series until removing one more would break the claim. Two to four is typical. Grey out the rest if you must keep them.
- Annotate the gap. Label the difference directly on the chart ("+38%", "below the 65°F threshold"). Don't make readers compute it.
- Strip chartjunk. Gridlines lighter or gone, no 3D, no gradients, no borders, legends replaced by direct labels where possible. Wikipedia's overview of chartjunk is a quick reference for what counts.
- Keep the rigour in the title and the footnote. Sample size, date range, and data source belong there. Simplifying the visual is not an excuse to hide the methodology.
When complexity is justified
The simplicity rule has honest exceptions.
Memorability. A 2010 study by Bateman and colleagues compared embellished charts with minimalist versions and found that decoration didn't hurt comprehension and slightly improved long-term recall. If your goal is to be remembered a month later rather than to win a decision today, some ornament can earn its place.
Exploration. When the audience's job is to find patterns rather than evaluate one, density is a feature. Scientific figures and analyst dashboards live here.
Expert audiences. Readers who use a chart type daily (epidemiologists with log-scale curves, traders with candlesticks) decode "complex" forms as fluently as you decode a bar chart.
The question to ask is not "is this chart simple?" but "is this chart the simplest one that makes my claim to this audience?"
Simple is not the same as honest
Everything so far argues for removing things. That advice carries an obvious risk: the easiest way to make a chart "cleaner" is to remove the parts that complicate your story. Simplification done badly isn't clarity; it's cherry-picking with good typography.
The traps are well known, and readers increasingly recognise them:
- Truncated y-axes. Starting a bar chart at 80 instead of zero can make a 3% difference look like a cliff. Bars encode length; length starts at zero. If a small difference matters, use a dot plot or show the percentage change directly rather than inflating bars.
- Convenient time windows. A line that rises from January to June may have fallen from the previous July. Choose the range that fits the question, not the one that fits the answer, and state it in the subtitle.
- Dropped categories that change the picture. Removing ten of twelve channels (as above) is fine when the claim is about two of them. It's not fine when the claim is "we're growing" and the ten you dropped were shrinking faster than the two you kept grew.
- Aggregates that hide a split. An average that goes up while the median goes down is a different story, not a simpler one.
A useful test: would a sceptical colleague who saw the full dataset say your simple chart is a fair summary, or a selective one? If you're not sure, show them. The strength of a simple chart comes from the reader's confidence that nothing important was hidden. Lose that once and every clean chart you show afterwards gets read as spin.
Simplicity earns trust only when it's paired with disclosure: the full data available, the methodology in the footnote, and the choices you made visible to anyone who looks.
Takeaways
- Readers decode position and length far more accurately than angle, area, or colour. Simple charts use the accurate encodings.
- An argument is one claim. A chart that makes several claims makes none well.
- A chart title should state the verdict, not the topic.
- The Challenger engineers had the right data and lost the argument because temperature was never put on an axis.
- Nightingale's rose persuaded despite its encoding, because the effect was enormous. Most effects aren't.
- Dashboards report; single charts argue. Know which you're building.
- Simplicity earns trust only when the full data is one click away.
- Simplify the visual, never the methodology. Rigour moves to the title and footnote; it doesn't disappear.
The next time you're tempted to add a series, an axis, or a gradient, ask what it costs the reader. Usually the answer is: the two seconds in which they would otherwise have agreed with you.
