cleardatavizTHE WORK,
step by step
Scroll through three real redesigns. The chart on the left changes along with the text, so you can see exactly which choice has which effect. These are the three examples I run into most often, and where I think things could be better, different, or clearer. Making the right choice isn’t always easy. It starts with three questions. Which question do you want the visualization to answer? Who are you making it for? And are you following the guidelines for data visualization?
↓ SCROLL
CASE I THE SAME DATA, EIGHT DIFFERENT CONCLUSIONS
COFFEE SALES · KILOGRAMS AND REVENUEOur revenue follows the number of kilograms sold exactly
LEFT KILOGRAMS 30–170 · RIGHT REVENUE €200–1,600
Two lines, two different measuring sticks.
You’re asked to chart coffee sales for the last five months, in kilograms and in euros. You have daily data available. Both measures suit a line chart well. Blue shows the kilograms of coffee sold, green shows the revenue in euros. Blue belongs to the numbers on the left, green to the numbers on the right. That right axis runs from 200 to 1,600 euros. With that setting, the two lines sit almost on top of each other. The conclusion you write down: revenue tracks the number of kilograms exactly.
Revenue grows faster than the number of kilograms.
The title changes! What happened? Only the right measuring stick: it now starts at 0 instead of 200. That pushes the green line upward, and it now sits above blue for the entire period. New message: we’re getting more euros out of every kilogram. Nothing changed in the numbers.
More kilograms, no extra euros.
Now I stretch the right measuring stick to 6,400 euros. The green line gets flattened against the bottom and looks like it’s standing still, while blue shoots up. The message is the exact opposite of step 02: we’re selling more, but not earning anything extra for it. Same data.
Revenue catches up with volume.
Again, only the right measuring stick changes. Now green starts below blue and crosses the blue line in March. Everyone reads a crossing like that as a turning point: “something happened here.” But with two different measuring sticks, you decide for yourself where the lines cross. The turning point is invented.
Only now can you compare them fairly.
The second axis is gone. Both series have been converted to percentages and start at 100% in January. What you see is how much each series grew relative to its own starting point: revenue pulls ahead of kilograms. So: we didn’t sell much more coffee — we sold pricier coffee.
One line that answers the question directly.
Subtract the kilograms from the euros. The blue line becomes a flat line at 0 percent, and green shows how far revenue is ahead. In March that’s plus 47 percent, then it settles back to around plus 14 percent. One line, one question, one answer. There’s no second axis left to steer the picture. This approach looks simple, but pay attention: because every recognizable sales figure has been converted into a percentage, your audience may start to doubt it. If you choose this view, explain clearly what you did.
The easiest solution: two charts.
Pull the series apart. Here, just the kilograms, on an axis that belongs to kilograms. No second measuring stick, no legend, nothing to misread: from just over 70 kilograms to 120 in four weeks, peaking at 134, then settling back to 97.
Two simple visuals say enough.
The same treatment for revenue: its own axis, neatly starting at zero. Doubled in March, peaked at €1,330, now at €860 — about 35% below its peak. Placed side by side, these two charts tell the same story as steps 05 and 06, but you don’t need to explain anything alongside them. Steps 05 and 06 are more powerful, but ask your audience to follow you through indexing and difference math. So it’s a matter of choice: how much explanation suits this audience?
This isn’t just my opinion.
Stephen Few wrote the standard piece on dual axes back in 2008: scaling two series independently invites the reader to compare their heights, even though that comparison is meaningless. He can’t think of a single situation where a second axis works better than the alternatives — and names exactly the two you saw above: series in separate charts side by side, or converting to one scale with an index point. Jon Schwabish sums up the core problem: by choosing the axis bounds, you can make two series look as correlated as you like. And Cole Nussbaumer Knaflic warns about exactly step 04: a crossing looks like news, but it only follows from the chosen scale.
Lying with data, or letting the data tell the story?
All nine visuals above come from a single file. Not one number was changed, not one row was left out. The only difference is what I chose: which axis, which bounds, which title. You always make that choice — even when you don’t think about it. That’s exactly why no neutral chart exists: every visualization is a decision about what the reader sees, and therefore concludes. Ask yourself, for every visual: does this choice help my reader understand reality, or does it help me make my point? Your credibility rests on that difference — and you only lose it once.
Our revenue follows the number of kilograms sold exactly
LEFT KILOGRAMS 30–170 · RIGHT REVENUE €200–1,600
Two lines, two different measuring sticks.
You’re asked to chart coffee sales for the last five months, in kilograms and in euros. You have daily data available. Both measures suit a line chart well. Blue shows the kilograms of coffee sold, green shows the revenue in euros. Blue belongs to the numbers on the left, green to the numbers on the right. That right axis runs from 200 to 1,600 euros. With that setting, the two lines sit almost on top of each other. The conclusion you write down: revenue tracks the number of kilograms exactly.
Revenue grows faster than the number of kilograms
RIGHT €0–1,400 — THE AXIS NOW STARTS AT ZERO
Revenue grows faster than the number of kilograms.
The title changes! What happened? Only the right measuring stick: it now starts at 0 instead of 200. That pushes the green line upward, and it now sits above blue for the entire period. New message: we’re getting more euros out of every kilogram. Nothing changed in the numbers.
More kilograms, no extra euros!
RIGHT €400–6,400 — THE AXIS HAS BEEN STRETCHED
More kilograms, no extra euros.
Now I stretch the right measuring stick to 6,400 euros. The green line gets flattened against the bottom and looks like it’s standing still, while blue shoots up. The message is the exact opposite of step 02: we’re selling more, but not earning anything extra for it. Same data.
Revenue catches up with volume
RIGHT €500–1,400 — THE AXIS HAS BEEN COMPRESSED
Revenue catches up with volume.
Again, only the right measuring stick changes. Now green starts below blue and crosses the blue line in March. Everyone reads a crossing like that as a turning point: “something happened here.” But with two different measuring sticks, you decide for yourself where the lines cross. The turning point is invented.
Not more coffee sold — just pricier coffee
ONE SCALE · BOTH SERIES INDEXED TO JANUARY = 100%
Only now can you compare them fairly.
The second axis is gone. Both series have been converted to percentages and start at 100% in January. What you see is how much each series grew relative to its own starting point: revenue pulls ahead of kilograms. So: we didn’t sell much more coffee — we sold pricier coffee.
In March every kilogram earned 47% more — now just 14%
REVENUE MINUS KILOGRAMS — ONE LINE THAT SHOWS THE DIFFERENCE ITSELF
One line that answers the question directly.
Subtract the kilograms from the euros. The blue line becomes a flat line at 0 percent, and green shows how far revenue is ahead. In March that’s plus 47 percent, then it settles back to around plus 14 percent. One line, one question, one answer. There’s no second axis left to steer the picture. This approach looks simple, but pay attention: because every recognizable sales figure has been converted into a percentage, your audience may start to doubt it. If you choose this view, explain clearly what you did.
From 70 to 120 kilograms in four weeks
KILOGRAMS ONLY · OWN AXIS, NO SECOND SCALE
The easiest solution: two charts.
Pull the series apart. Here, just the kilograms, on an axis that belongs to kilograms. No second measuring stick, no legend, nothing to misread: from just over 70 kilograms to 120 in four weeks, peaking at 134, then settling back to 97.
The revenue doubled in March and now sits 35% below its peak
REVENUE ONLY · AXIS FROM ZERO
Two simple visuals say enough.
The same treatment for revenue: its own axis, neatly starting at zero. Doubled in March, peaked at €1,330, now at €860 — about 35% below its peak. Placed side by side, these two charts tell the same story as steps 05 and 06, but you don’t need to explain anything alongside them. Steps 05 and 06 are more powerful, but ask your audience to follow you through indexing and difference math. So it’s a matter of choice: how much explanation suits this audience?
Indexing or splitting — two routes the experts recommend
STEPHEN FEW (2008) · JON SCHWABISH (2022) · COLE NUSSBAUMER KNAFLIC (2016)
This isn’t just my opinion.
Stephen Few wrote the standard piece on dual axes back in 2008: scaling two series independently invites the reader to compare their heights, even though that comparison is meaningless. He can’t think of a single situation where a second axis works better than the alternatives — and names exactly the two you saw above: series in separate charts side by side, or converting to one scale with an index point. Jon Schwabish sums up the core problem: by choosing the axis bounds, you can make two series look as correlated as you like. And Cole Nussbaumer Knaflic warns about exactly step 04: a crossing looks like news, but it only follows from the chosen scale.
Lying with data, or letting the data tell the story?
ONE DATASET · NINE VISUALS · NINE CONCLUSIONS — THE CHOICE BELONGS TO THE MAKER
Lying with data, or letting the data tell the story?
All nine visuals above come from a single file. Not one number was changed, not one row was left out. The only difference is what I chose: which axis, which bounds, which title. You always make that choice — even when you don’t think about it. That’s exactly why no neutral chart exists: every visualization is a decision about what the reader sees, and therefore concludes. Ask yourself, for every visual: does this choice help my reader understand reality, or does it help me make my point? Your credibility rests on that difference — and you only lose it once.
CASE II THE AXIS THAT DOESN’T START AT ZERO
SUNSHINE HOURS PER YEAR · TOP 10AXIS FROM 1,850 TO 2,250 — THE DIFFERENCE LOOKS HUGE
Ten years, huge differences.
The ten sunniest years, sorted from most to least sunshine hours. 2025 is marked in dark blue. What you see: 2022 has pulled far ahead and 2009 barely registers. Now look at the first number in the bottom left of the axis.
The shape says 28%, the numbers say 5%.
This chart’s axis doesn’t start at zero — it starts at 1,850. And that’s the problem. In a bar chart, the length of the bar tells the story. You see the effect immediately. 2022 looks almost a third longer than 2025. But the real difference between 2,235 and 2,125 sunshine hours is under five percent. This chart’s axis doesn’t start at zero — it starts at 1,850. And that’s the problem. If you don’t start that axis at zero, the proportions no longer hold and you show a different story.
With the axis at zero, it’s suddenly boring.
Same ten years, now drawn from zero with the values shown in the bar. Everything sits between 1,890 and 2,235 sunshine hours: ten nearly equal bars. Boring is the truth here — and nobody needs to estimate anymore, because the number is right there.
Seven of the ten sunniest years are recent.
Now that the fake differences are gone, the real pattern stands out: only 1959 and 2003 come from a different era — the rest are from the last ten years. Grey out those two years and put the conclusion in the title, and the image does its job at a single glance. That’s the payoff of an honest axis: you only find something once you stop hiding it.
The field agrees on this point.
Andy Kirk explains why it goes wrong: you read a bar chart by judging the absolute length of the bar — cut that off, and you read something other than what’s actually there. Stephanie Evergreen is absolute about it: for bars, the axis must always start at zero, because bars encode their value in length; shorten it, and comparison becomes misleading. Datawrapper doesn’t even let you set the starting point, citing research: readers of a truncated chart take away an exaggerated version of the message. Alberto Cairo does leave room for a different baseline on line charts — but not on bars, and on that last point the sources agree.
Your tool picks the starting point. You’re responsible.
BI and other software tools often pick a "nice" starting point themselves as soon as the values sit close together. There’s no warning, no break in the axis — just a chart that looks polished and exaggerates the difference. That makes this the most common mistake in real-world reporting, and almost never out of bad intent. Hence the same test question as in the previous example: does this choice help my reader see reality, or does it help me make my point?
AXIS FROM 1,850 TO 2,250 — THE DIFFERENCE LOOKS HUGE
Ten years, huge differences.
The ten sunniest years, sorted from most to least sunshine hours. 2025 is marked in dark blue. What you see: 2022 has pulled far ahead and 2009 barely registers. Now look at the first number in the bottom left of the axis.
AXIS FROM 1,850 TO 2,250 · REAL DIFFERENCE 2022 VS 2025: 5%
The shape says 28%, the numbers say 5%.
This chart’s axis doesn’t start at zero — it starts at 1,850. And that’s the problem. In a bar chart, the length of the bar tells the story. You see the effect immediately. 2022 looks almost a third longer than 2025. But the real difference between 2,235 and 2,125 sunshine hours is under five percent. This chart’s axis doesn’t start at zero — it starts at 1,850. And that’s the problem. If you don’t start that axis at zero, the proportions no longer hold and you show a different story.
AXIS FROM ZERO · VALUES SHOWN IN THE BAR
With the axis at zero, it’s suddenly boring.
Same ten years, now drawn from zero with the values shown in the bar. Everything sits between 1,890 and 2,235 sunshine hours: ten nearly equal bars. Boring is the truth here — and nobody needs to estimate anymore, because the number is right there.
Seven of the ten sunniest years fall after 2015
AXIS FROM ZERO · 1959 AND 2003 IN GREY — THE REST IS FROM AFTER 2015
Seven of the ten sunniest years are recent.
Now that the fake differences are gone, the real pattern stands out: only 1959 and 2003 come from a different era — the rest are from the last ten years. Grey out those two years and put the conclusion in the title, and the image does its job at a single glance. That’s the payoff of an honest axis: you only find something once you stop hiding it.
With bars you read length — so the axis starts at zero
SOURCES: A. KIRK (NATUREJOBS) · S. EVERGREEN · DATAWRAPPER ACADEMY · A. CAIRO
The field agrees on this point.
Andy Kirk explains why it goes wrong: you read a bar chart by judging the absolute length of the bar — cut that off, and you read something other than what’s actually there. Stephanie Evergreen is absolute about it: for bars, the axis must always start at zero, because bars encode their value in length; shorten it, and comparison becomes misleading. Datawrapper doesn’t even let you set the starting point, citing research: readers of a truncated chart take away an exaggerated version of the message. Alberto Cairo does leave room for a different baseline on line charts — but not on bars, and on that last point the sources agree.
Your tool picks the starting point. You’re responsible.
ONE DATASET · TWO VISUALS · TWO CONCLUSIONS — THE CHOICE BELONGS TO THE MAKER
Your tool picks the starting point. You’re responsible.
BI and other software tools often pick a "nice" starting point themselves as soon as the values sit close together. There’s no warning, no break in the axis — just a chart that looks polished and exaggerates the difference. That makes this the most common mistake in real-world reporting, and almost never out of bad intent. Hence the same test question as in the previous example: does this choice help my reader see reality, or does it help me make my point?
CASE III TWO CIRCLES YOU CAN’T COMPARE
COFFEE SALES BY TYPE · THIS YEAR VS LAST YEARFIVE COLORS · LEGEND BELOW · NO VALUES
Two circles, five colors, one legend.
This year on the left, last year on the right, five coffee types per circle. The question is: what changed? Go ahead and try. You have to look up each type’s color in the legend, then compare one slice with another — and people are bad at comparing pie slices, because you have to estimate angles.
With the numbers in, you start calculating.
Now the counts are shown in the slices. Handy, but notice what happens: you’re no longer looking, you’re calculating. Espresso went from 119 to 101, Macchiato from 25 to 40. The chart doesn’t read that out for you — you do, in your head.
Five colors that add nothing.
I give every slice the same shade. What did you lose? Nothing, because the colors didn’t stand for anything — they only distinguished names, and you can put those names right there too. Color is too precious to waste on labels.
Name, count, and share right on the slice.
The legend can go: every slice carries its own name, count, and percentage. Now your eye no longer has to go back and forth. It’s still hard to compare between the two circles — the slices sit in different places and the totals differ.
Highlight what the conversation is about.
Macchiato is the story: from 25 to 40, from 7% to 14% of sales. Marked in dark blue, it stands out in both circles. This is the most you can get out of a pie chart — and the reader still has to lay two visuals side by side.
Want to show change? Connect the dots.
Same numbers, one visual: this year on the left, last year on the right, one line per type. Now you see it at a single glance. Espresso, Cappucino, and Latte trend down; Macchiato and Lungo trend up — Macchiato the strongest. No estimating angles, no legend, no mental math. A pie chart answers the question "how is it distributed?" This question was "what changed?" — and that calls for a different visual.
FIVE COLORS · LEGEND BELOW · NO VALUES
Two circles, five colors, one legend.
This year on the left, last year on the right, five coffee types per circle. The question is: what changed? Go ahead and try. You have to look up each type’s color in the legend, then compare one slice with another — and people are bad at comparing pie slices, because you have to estimate angles.
VALUES SHOWN IN THE SLICES — YOU START CALCULATING INSTEAD OF LOOKING
With the numbers in, you start calculating.
Now the counts are shown in the slices. Handy, but notice what happens: you’re no longer looking, you’re calculating. Espresso went from 119 to 101, Macchiato from 25 to 40. The chart doesn’t read that out for you — you do, in your head.
ONE SHADE — THE COLORS ADDED NOTHING
Five colors that add nothing.
I give every slice the same shade. What did you lose? Nothing, because the colors didn’t stand for anything — they only distinguished names, and you can put those names right there too. Color is too precious to waste on labels.
LABELS ON THE SLICE · LEGEND NO LONGER NEEDED
Name, count, and share right on the slice.
The legend can go: every slice carries its own name, count, and percentage. Now your eye no longer has to go back and forth. It’s still hard to compare between the two circles — the slices sit in different places and the totals differ.
MACCHIATO HIGHLIGHTED: FROM 7% TO 14% OF SALES
Highlight what the conversation is about.
Macchiato is the story: from 25 to 40, from 7% to 14% of sales. Marked in dark blue, it stands out in both circles. This is the most you can get out of a pie chart — and the reader still has to lay two visuals side by side.
Macchiato grows fast, Espresso and Latte decline
ONE VISUAL · ONE LINE PER TYPE · NO MORE ESTIMATING ANGLES
Want to show change? Connect the dots.
Same numbers, one visual: this year on the left, last year on the right, one line per type. Now you see it at a single glance. Espresso, Cappucino, and Latte trend down; Macchiato and Lungo trend up — Macchiato the strongest. No estimating angles, no legend, no mental math. A pie chart answers the question "how is it distributed?" This question was "what changed?" — and that calls for a different visual.