Finetune your graphs by discussing this question with your colleagues: should our graph’s axes extend to the dataset’s actual maximum value or to the potential maximum value?
Option A: Axis stretches just past the actual maximum value (30% in this example)
The biggest number is 26%, and the axis goes from 0% to 30%, just past the biggest number in the bunch. Your stakeholders will think the 26% looks enormous because it stretches all the way across the screen. Wow, look at our numbers! Great news! Of course, we’ll have to work on Category D, but we can certainly improve that number! Especially when the other categories are looking so good! Talk to your teammates: is this the message we’re going for?
Option B: Axis stretches all the way to the potential maximum value (100% in this example)
What if the numbers have the potential to stretch all the way to 100%? (The percentage of attendees who said they’d recommend your conference to a colleague, the percentage of students who graduated on time, and so on.) If you’re trying to hit 100%, now the 26% isn’t looking so hot. Argh. We thought everything was going so well! We’ve got so much to improve upon. Where do we even start?! Ask your teammates: is this the right time to risk overwhelming the stakeholders with so much bad news?
My advice: Good facilitation skills are an ingredient of good data visualization. I often begin with the first graph to avoid scaring stakeholders away from action. As we get to know each other — and they get to know their numbers — I slowly introduce the second style.
When choosing minimum and maximum axis values, what other factors do you consider?
Download the template.
8 Comments
Leave a Reply
The Graph's Scale: Actual Maximum Value or Potential Maximum Value?
8 Comments

But, Ann, isn’t that first graph just plain dishonest? I know Tufte would say that, and Cairo comes close to that in his new book.

I don’t think the first graph is dishonest. If the axis didn’t start at zero, then yes, it would be dishonest. If it started at 8% for example, then category D would be disproportionally smaller than the rest (all the categories would be disproportionate to their true values). But since the axis starts at zero, having it end anywhere below the potential maximum, simply makes all the categories larger, but proportionally correct. i.e. in both examples above, category A is about 3 times larger than category D. The proportion of the bars doesn’t change, just the size.

It could be argued that for a percentage figure, the proportions of the white space are just as important, in which case the first graph is dishonest.
Fundamentally though I think it’s a question of context. If 100% is a realistic target / expectation (eg election turnout), then having anything other than a “full” axis is dishonest, while if it’s always something that will be lower values (eg election results by party) then a proportionate axis would seem entirely legitimate.
Even when 100% is still achievable, as Ann says we can control the message by changing the axis from 30% (against which 26% looks TOO good) to 50%, so that we can focus attention on the 26% without creating depression all around. It doesn’t have to be 30% or 100% – there are points in between!


[…] I’ve been showing you one tip at a time, like deciding how far you’ll stretch your scale, whether you’ll use a regular or stacked chart, and whether your chart will be oriented […]
But, Ann, isn’t that first graph just plain dishonest? I know Tufte would say that, and Cairo comes close to that in his new book.
I don’t think so. Here’s an example from Pew where they displayed percentage point differences, and the axis stretches from 0% to around 10% (just past the biggest number, which is 9%). There could potentially be a 100% percentage point difference so the axis could stretch from 0 to 100%, but doesn’t. https://twitter.com/pewinternet/status/793817691589148672
This example uses units (debt in trillions), not percentages, but it’s another Option A example. The axis stretches just past the biggest number. https://twitter.com/acotgreave/status/790539189863477248
Last one. From the Economist. The percentage of scientific papers with spreadsheet errors. Axis stretches from 0% to 40% (just past the biggest number), even though it’s possible that all 100% of papers could’ve had spreadsheet errors. https://twitter.com/TheEconomist/status/773713844573237249 I’m sure some people choose to use axis minimums and maximums to be dishonest, but that’s not what I’m advocating.
I don’t think the first graph is dishonest. If the axis didn’t start at zero, then yes, it would be dishonest. If it started at 8% for example, then category D would be disproportionally smaller than the rest (all the categories would be disproportionate to their true values). But since the axis starts at zero, having it end anywhere below the potential maximum, simply makes all the categories larger, but proportionally correct. i.e. in both examples above, category A is about 3 times larger than category D. The proportion of the bars doesn’t change, just the size.
It could be argued that for a percentage figure, the proportions of the white space are just as important, in which case the first graph is dishonest.
Fundamentally though I think it’s a question of context. If 100% is a realistic target / expectation (eg election turnout), then having anything other than a “full” axis is dishonest, while if it’s always something that will be lower values (eg election results by party) then a proportionate axis would seem entirely legitimate.
Even when 100% is still achievable, as Ann says we can control the message by changing the axis from 30% (against which 26% looks TOO good) to 50%, so that we can focus attention on the 26% without creating depression all around. It doesn’t have to be 30% or 100% – there are points in between!
Andy – Great logic. I’m trying to get more people to think carefully like you’re doing here rather than just going with whatever minimum and maximum values that their software program randomly gives them.
[…] I’ve been showing you one tip at a time, like deciding how far you’ll stretch your scale, whether you’ll use a regular or stacked chart, and whether your chart will be oriented […]