1984 may have been the year of the nerd. I didn’t know it at the time – I had not grown into my nerdiness yet – but 1984 was a year that shaped my future.
Some pretty awesome stuff for nerds came out in 1984. Tetris, Alex Trebek on Jeopardy, the Apple Macintosh, Revenge of the Nerds, and a seminal paper titled “Graphical perception: theory, experimentation, and application to the development of graphical methods.”
Yep, 1984 was a great year for nerds.
The paper was written by William Cleveland and Robert McGill, two statisticians at AT&T Bell Labs. In the paper, they propose basic guidelines for choosing the best way to encode data into a graphic that enables the most accurate comparisons, based on how humans perceive things. Since comparison is probably the most common use case for a chart, their framework is very useful.
The graphic below is adapted from their work and many other references to it in the past 30 years. From left to right, you see a spectrum of data encoding methods you can use. As you move right, Cleveland and McGill argue that humans are able to more accurately make comparisons.
Source: created by GetNerdyHR, adapted from Cleveland and McGill 1984 and Alberto Cairo’s The Functional Art
This becomes a super helpful tool when choosing chart types, and makes clear why bar charts are so helpful.
Let’s illustrate the framework with an example, borrowed proudly from Alberto Cairo’s The Functional Art (a fantastic read).
Based on the shading of the boxes below, try and rank order them from smallest to largest, where a darker color is analogous to “largest” data.
Most people can tell that B is large. Most people, though, would struggle to compare C and F. Your eyes have to jump back and forth several times and try to focus on an attribute that is not natural. It’s also impossible to gauge how much larger E is than A. It is larger, but there is no scaling to it.
Now let’s move to the right on the Cleveland and McGill’s framework. Repeat the rank order exercise using bubbles, where a larger area of the bubble represents “larger” value.
A little easier? Maybe. But still challenging. You can get a sense for how much larger E is than A, but most of you will actually draw an incorrect conclusion. That’s because most people will compare the diameter (width) of the circle, but the data was actually encoded based on the area of the circle. Think back to elementary school (maybe even back to 1984 for some).The area of the circle is equal to half the diameter (radius), squared, times Pi. So unless you are great at taking the square root of a visual concept and then dividing by Pi… you don’t really have a clue how much larger one circle is from another.
So let’s make it less challenging and jump all the way to comparing a position along an aligned scale. Yep, essentially a stripped down bar chart.
I imagine you can now compare quite accurately. Even without axes, gridlines, data labels, or any other trick to make the comparison obvious, it’s much easier to compare the relative size of each item. B is indeed largest, and it’s easy to spot that E is larger than A. You can also get a sense of how much larger E is than A without using any math.
This is what makes bar charts so practical for communicating data. They may be boring, but people get it. Immediately. We should always be asking ourselves if there is a simpler way to communicate our message.
It should be stated that this does not suggest there isn’t a very useful place for other forms of encoding data. A nice Choropleth map is a great way to give a reader a feel for a geographic data set, for example. Cleveland and McGill’s advice pertains to accurate comparisons, not an overall understanding of a data set. There is a time and place for a variety of visualization formats. Cleveland and McGill’s now 33-year-old model provides some understanding of why a bar chart is often the best answer.
Now you have one more reason to look back on 1984 fondly. You are welcome, by the way for the links to online Tetris, Jeopardy, and the Mac emulator. Childhood, revisited.