Why Pie charts are evil?

  • 31 January 2022
  • 4 replies
  • 144 views

Userlevel 2
Badge
  • Ataccamer
  • 2 replies

Data Stories provides a wide range of charts, it includes also Pie and Donut ones, even though I do not recommend using them, and the same applies to a Gauge chart. Let's look at the reasoning why and why I'd recommend using a (stacked) bar chart instead.


Space usage

In data visualization (reporting, data storytelling) you have limited space you can use. Your task is to fill the limited space with the most value for the end viewer/user. Note, most value does not necessarily mean most charts, sometimes one curated chart is far better than five. 

The problem with Pie, Donut and Gauge charts is that they use more space in comparison to an alternative that provides the same value - a bar chart. 

Sales distribution on a Pie Chart

More space is used due to 2 reasons:

  • Pie and Donut charts use Area to give value to a single category. Bar chars use just distance. An Area just takes more space
  • All these charts are circular, which creates unnecessary space around the chart object

Bellow find a better alternative for the chart above

Sales distribution on a Bar Chart [$]

Readability and comparability

The human eye and mind are good at estimating and comparing a distance. However we are not that good at doing the same with the area, it's just more complex as another dimension comes in place.

  • As described above Pie and Donut charts use Area (2 dimensions) to give importance to a single category/value, bar chars use just distance (1 dimension). Gauge chart uses distance but plots it in an area
  • When categories/values are packed together in these 3 charts, it's not that easy to compare by how much (or how many times) is the second one bigger, not to mention that if there are a lot of categories/values sometimes you can’t even tell which category/value is bigger than the other one.
Harvest results on a Pie chart

Common concerns

Sometimes people recommend using a pie chart when you want to display percentages and/or a few values. I believe even for percentages and few values bar charts are a better alternative, see charts bellow

Yes/No/Maybe on a Pie Chart

What chart/s to use instead?

To substitute the use of those 3 charts I recommend using a bar chart - but which type to use?
There is no single rule as it solely depends on your use case, data and space you have. 

When you have a few values stacked chart might be a good pick. When you have a lot of values it makes sense to use just a classical bar chart to allow users a better value comparison

Yes/No/Maybe on 2 Bar charts

   

Another reason to select any of the 3 charts is to just use something else as your dashboard or a story is full of bar charts and you just want some different visualization. This is a common issue that is not very often verbalized. I believe using that using the correct tooling is far more important. Also, don't be afraid of boringness, its the other way around, the reporting is then more consistent

Visualization enrichment

Bar charts also allow a much bigger range of options and additional data points to be introduced, enriching the visualization while having the same area. 

Probably the best example is an implementation of them called Bullet charts, these provide options to show a comparison to targets and ranges of performance

Revenue on a Bullet chart [$]

Moreover, the evident benefit of bar chars is their possibility for sorting their categories from largest to smallest

Note that introducing any 3D visualization (created of their 2D version without any additional data points) does not bring additional value, it just increases complexity - another dimension to take into consideration when looking at graph values/categories

Sports preference on a 3D chart

Exceptions

No rule or recommendation is carved to stone and sometimes using Pie, Donut or Gauge charts is suitable, one of the examples is gauges in cars

Some of the reasons for using these charts may include:

  •  We have been left with a square space for visualization
  •  Users are used to this chart from past 
  •  It would be complicated to create the chart as a bar (i.e. analogue measuring devices)
  •  Data are in % and only two values are displayed - this allows the user to see data are percental and user can quickly see proportion to 50%
  • Additional data points are added, like in the doughnut chart when additional data are put in its middle

4 replies

Userlevel 1

I agree that pie charts are evil, but (and there is always a but..) when putting together a data story be it a report, visualisation or dashboard it is important to consider who the audience is. While pie charts are evil and there are generally better visualisation options available, if your intended audience is expecting/wanting a pie chart then it there may be a case for including them.
I have seen end users reject a dashboard as they were so used to seeing a certain dataset visualised as a pie that they automatically rejected any other form.

If your audience is resistant to change, then a pie chart may be a necessary evil.

Regarding the donuts, I believe that they use (arc) length to represent values, not area. Pie charts represent values through the use of angles which is not a good option as humans are not wired to accurately perceive differences in angles.

If a pie chart is needed, it is good to follow some conventions, such as ordering each segment from largest to smallest starting at the 12 o’clock position.

And definitely 3D charts are almost always never a good option.

Userlevel 2
Badge

Hi Daniel, 

About the audience: I agree with you. I would dare to add that it's also the job of BI analysts/developers to explain and educate the stakeholders (sure, not always they have a chance for that)

About Donuts: I guess that depends on the implementation of how values are represented, but angles are definitely used (by definition) and they are indeed not good. I always perceived the donuts chart as a pie chart with a hole :)

About Pie chart conventions: Again, I absolutely agree. Also if possible I also suggest sorting other charts by some logic, if that is not in contradiction with the business view

Userlevel 1

There is growing appreciation of the need for improved data literacy across the entire organisation spectrum, from the C-suite to the customer facing. As to who should take the reins in driving this education and literacy uplift within an organisation, that is debatable and from my experience it is more about individuals’ skillsets. 

There may be some BI analysts/developers who are able to assist in the education and training of others within their organisation, but to expect this of any and all BI analysts/developers is foolhardy. It is well worth looking across the organisation, identify data literacy leaders and potential advocates no matter where they sit and supporting them to bring about the cultural change to embed mature organisational data literacy.

A good BI analyst/developer may not be able to transition over to an education/advocacy role. For those that are able to, support them, celebrate them. The greatest challenge though is to shift the organisational mindset that data literacy is the realm of the BI analyst/developer, the pocket of data people within the IT/Tech divisions. Data literacy, just like language-literacy, is essentially in all roles, be it the creator of a dashboard or a consumer of one. Be it those generating the raw data, or those refining and value-adding throughout the data’s lifecycle. 

Userlevel 2
Badge

Daniel, thank you for your comment/contribution, I simply love your response and have nothing to add :)

 

 

Reply