Here is a simple but vital charting rule.
Start your bar (or column) charts from zero.
To illustrate why you should do this, let me share a personal example.
Over the weekend, the Jon Peltier visited Wellington. He is staying with Jeff (who occasionally guest blogs on Chandoo.org). On Sunday, we all decided to hike up a small mountain near my house for a leisurely family picnic.
While on the top of the mountain, Jo (my wife) took a few pics of us three Excel geeks. As we were standing on a sloping mountain face this is how the pictures look:
Looking at the picture on left, you would confidently say that I am way shorter than other two. But picture on right tells a different story.
Of course, the reality is somewhere between two pictures. It is difficult to conclude who is tall, who is short just looking at the pics simply because the baseline is sloping.
But we can’t have sloping baselines in Excel charts
You are right we can’t. But we can still confuse people with an arbitrary axis start. Like below:
The fix? always start your axis at zero for bar (or column) charts:
Simple. Set the axis start point to zero (Select axis, press Ctrl+1, and from Axis options set minimum to 0).
So there you go. The shortest Excel charting tip ever, but still stands tall when it comes to telling great stories.
More charting principles:
If you are in the mood for some more charting theory and elegant methods, check out below links.
25 Responses to “Start your bar charts from zero – Excel geeks screaming at you from mountain top”
Great illustration to drive home the point. Makes me lmao.
Then again coming from a math/statistic background, statisticians sometimes do purposely start with a non-zero axis or log axis to "provide" a visual trend to what they want to convey.
No self-respecting statistician would plot a bar chart on a log axis, nor a bar chart with a non-zero bseline.
Many statisticians will lose their self-respect if the marketing guy has them in a head-lock and is waving dollar bills in their face.
Haha. I'm like two times louder than you, but you have two times the amount of hair on your head than I do!
How many Excel geeks does it take to teach a simple lesson?
I had aa awesome stay in Wellington, and it was great meeting Chandoo for the first time and catching up with Jeff.
Too much Excel power on that hill for sure! But what a view it is.
An Excel summit!
Brilliant, David! I'll be using that tonight at the meetup when I introduce Jon.
Good one, David.
Perfect description! 🙂
Like many things in life...it depends. The example given is to show differences in height. So the non-zero base causes a misleading visual. However, if you are tracking the growth rate of secured loan pools across years by month and want to show a rolling 13 month view, then a non-zero base would work just fine. Specifically if the loan pools are growing by a positive rate over time. Some would say use a line. Except it is useful to also compare the month to month growth rate using the same chart. This allows the differences in rate to 'pop out' for the viewer (considering the y axis in tens of thousands of units).
What is the purpose of charts? Why, to convey information, of course. If they don't do their job then they are wrong.
I read Darrell Huff almost 25 years ago and of course I always look for the axis when I see a chart, so naturally I have an aversion for non-zero starts.
However, there are many, many variables where a miniscule change has a great significance - think blood values, perhaps, or financial figures. If I'm presenting to our board of directors and use a zero-start axis all I'm showing them is a series of very long, equal bars. Amount and value of information given: zero, like the beginning of the axis.
So: in essence, I agree. But the technique absolutely does have its purpose.
Did you notice the title of the article? "Start your BAR charts from zero". So chart type is an important aspect of this discussion.
The value of a bar is encoded by the length of the bar, and a comparison of the bar's length to other bars' lengths. If you cut off the bottom of the bar, then the encoding is incorrect, and no amount of conscious thought about the axis scale will overcome the eye's and brain's precognitive judgement of the value of the bar.
The value of a marker is encoded by the position of the marker along a scale, and a comparison of the marker's position to other markers' positions. If you need to show small differences, either plot actual deltas, or plot the values in a marker-containing chart (line or scatter).
While I agree with what you say, the point remains that such extreme bars convey no information whatsoever. So something must be done. Bias, sure, but it cuts both way - perceiving them the same when the difference is huge in significance is also a bias.
Now deltas is a great idea indeed and much closer to the spirit as it should be - you could even add labels for absolute values. However, this is not foolproof either.
Consider this: my corporate site may load at, say, five seconds. Flutter may indicate a problem, though that flutter is small. Plotting deltas, if a small scale is used, provides the information needed. However, for various reasons, a much bigger delta may be introduced, pushing all the other data points much closer together. So back at square one, hacking away at it trying to make it show something real...
(can't edit previous post...)
Same with blood sample values - there are indicators that need to be closely monitored and where significant deltas can be very small, but where you can also have huge spikes. Same thing.
So stop talking about "extreme bars", and stop plotting bars at all.
Oh gee, same things apply to the other kinds of charts as well, who would've imagined!
I don't understand what you're saying.
very nice… i really like your blog…
Really I'm Very Thankful to You, n I like The Posts in The way You Write/ Compose Keep Going On All The Best !!!
The Excel version of Mount Rushmore?
This is a really nice illustration. We have a very popular nature show on TV here called "Springwatch" and the other day the presenter showed a graph that seemed to show that British Isles farmland birds were totally extinct. They have actually decreased by about 42% IIRC, but it's a bit naughty to start the y-axis at 58%
Nice example ?
Nice example 😀
Horizontal numbers in line chart start at 1, so they consequentally get offset, reading 1, 11, 21, Jesus save us. Why is default not 0, 10, 20, 30. Including belittlement, no answers on the internet actually fix this issue for humans, math-minded or not.