Stop Using Pie-Charts

93 points by t_christensen 4 years ago | 32 comments

I would say there is a place for pie charts. The article seems to have accuracy as ultimate goal, if you need that accuracy; fully agree, pie chart isn't the best visualization. But, data visualization is often about story telling, and the small percentage differences don't always matter.

For example, if you are telling a story about how one of the slices is much bigger than the other ones, pie (and donut) charts, are a very effective and visually interesting way to tell that story. The other case where I like pie charts (and I do prefer donuts btw) is when the data isn't very accurate and "hiding" some of the detail is actually a better representation of reality.

So yeah, pick the chart that works best for the situation (and if possible, give multiple options), but I do not agree with writing off the whole chart type. Radar charts are more questionable imo haha.

Data visualization is in essence trading accuracy for readability, how much you want and need depends on the goal, audience and data.

jicea 4 years ago | |

I’ve seen an animation a few days ago [1] about the market share evolution of smartphone constructors for 20 years. The animated pie beautifully tells the raise of Apple, Samsung, Huawey and the fall of Nokia, RIM and others. So, fully agree, pie charts can be used if you don’t need the precision.

[1] https://www.linkedin.com/posts/jameseagle_mobilephones-datav...

Aliabid94 4 years ago | | |

Strong disagree on any type of animated chart. All the information in the animation could be represented in a single image as a multiseries line graph. Animation adds no value and spreads the information over time.

qiqitori 4 years ago | | |

Did they put pre-2000 Ericsson under Sony?

BeetleB 4 years ago | |

> For example, if you are telling a story about how one of the slices is much bigger than the other ones, pie (and donut) charts, are a very effective and visually interesting way to tell that story.

So is a stacked bar chart.

daveslash 4 years ago |

I had a college Statistics professor who would get triggered and go off on a (justified) rant every time he saw a 3D pie chart.

I am a big advocate or arranging the pie slices in order (largest to smallest, or smallest to largest).

I also really recommend The Wall Street Journal Guide to Information Graphics: The Dos and Don'ts of Presenting Data, Facts, and Figures https://www.amazon.com/Street-Journal-Guide-Information-Grap...

Spooky23 4 years ago |

The first example undermines the argument. It’s a demonstration that the author doesn’t understand a tool.

That pie chart tells me that the slices are approximately the same, which is a useful message to effectively deliver visually. Pie charts are great for understanding relative value.

Divining or comparing the precise values is not a good use for it.

edflsafoiewq 4 years ago | |

So does the second. It shows you almost immediately that about half the precipitation falls on the weekend, validating the claim, which is not at all clear from the bar chart.

Supermancho 4 years ago | |

It's less and less common to see pie charts without the specific %/values anyway. This is some sort of silly hangup by the author(s).

citizenpaul 4 years ago |

Pie charts are good for two things.

Showing how domanant one factor is.

Showing how no factor is dominant.

tgv 4 years ago | |

When used to visualize relative importance, it has the great advantage of an implied, absolute scale: you can see what's 50% without having to guess from the axis what the value is. You can even eyeball a majority, as in chart [1], although the colors of two adjacent parties happen to be very similar to my vision. You would not be able to do that with a bar chart such as [2]. The pie chart should not be your go-to chart, but it has its place.

[1] https://www.ft.com/__origami/service/image/v2/images/raw/htt...

[2] https://blogs.ft.com/the-world/files/2017/09/bar-chart.png

pie_flavor 4 years ago |

Pie charts let you tell percentages at a glance, which is a good thing to be able to tell. If your data is not dividing up a pie, a pie chart might not be the best fit, and if you need to tease out questions like "which is the biggest" given several close numbers, it definitely isn't a good fit. It especially helps when you order the slices in a hostile way - if it was otherwise confusing, just order them clockwise by size instead of going with the (irrelevant in this format) data order.

Any chart format has a set of data for which it's a terrible fit. Different formats have different information optimized to communicate at a glance, and if you're deliberately using the wrong one, you'll end up with a bad graph; that's not a takedown of the format itself.

BeetleB 4 years ago |

As someone who has fought this battle many times:

People don't want accuracy.

People don't want details.

People want pretty.

They will reject any other type of plot that is better than pie charts on the other two metrics.

Hence, pie chart.

The higher up you go, the more important the appearance is, and the less important the details are. The role of most presentations is not to get people to understand, but to impress. Senior folks have given feedback that "Your presentation slides don't have enough details. If you make it easy for the audience to understand, they will undervalue your work."

pattusk 4 years ago | |

Is it just me or does TFA's whole argument rests upon the assumption that you don't add labels to your pie charts ?

If I add percentages or values to the chart, there are no issues with either interpretation or comprehension. The pie chart is just a visually pleasant way to display data compared to, say, a table or a bar chart. Data that's displayed in a boring is quickly forgotten, but data presented in a striking way will help your point being remembered. It's akin to rhetoric/style in writing. You can write in a matter-of-fact descriptive way and bore 90% of readers to death, or you can articulate your point with striking metaphors, rhythm, etc... and make an impact.

This is an issue that I often notice with engineers. They assume that communication is transparent. A five page long table of figures? Sure! A front-end with tons of buttons, slide bars for every adjustable parameters and a full report of everything going on under the hood? Who wouldn't want that!

There's nothing wrong with pretty. You don't always have to sacrifice accuracy to get pretty. And pretty ensures that your accurate data isn't ignored.

BeetleB 4 years ago | | |

Again, it's a question of whether you are trying to convey details, or look pretty.

If you add numbers to the pie chart, then you don't need the chart at all. The benefit of a bar chart is that you don't need to list the individual percentages. At a quick glance, you can get more information from a bar chart than you can from a pie chart.

But if you're not trying to convey details, a pie chart is prettier.

jeffparsons 4 years ago | | |

If you want something like a pie chart, you can always go for a donut chart. It's still flawed as a communication tool if clarity is your main goal, but it's at least arguably prettier than a pie chart and easier to read the relative sizes accurately.

bitwize 4 years ago |

Pie charts are useful when there are large disparities between some of the data items. Like, here are our total expenses for last year, we spent 6% on administration, 14% on facilities, 11% on R&D, 33% on manufacturing, and 36% on sales and marketing. You can see at a glance, even with human eyesight's poor judgement of areas, what the dominating two expense areas are. Or, you know, here's a breakdown of the OS our dev team members use: 85% use Windows, 11% use macOS, 3% use Linux and 1% use "other".

Oftentimes they will be labelled as well with the exact percentage numbers which helps. It's not really a scientific visualization tool, but it adds punch to a presentation when you want to show that one or more subsegments of a whole really dominate the rest.

ralferoo 4 years ago |

This article is weird. It centres on the central premise that pie charts are useless because you can't tell exactly the relative sizes of slices that are very similarly sized. That's not the point of a pie chart, which is to show clearly which areas are big and which are insignificant.

The article then poses the question "Let’s See if it actually rains more on the weekend." Looking at the pie chart, it's immediately obvious at a glance, that YES, it does rain more at the weekend, as the two segments for Saturday and Sunday together account for nearly half the pie. Clearly the pie chart is perfect for answering this question. But then, the article launches into a whole spiel about how pie charts are useless because you can't tell which was rainier between Saturday or Sunday.

However, if you used an alternative presentation e.g. the lollipop charts as they suggest, then it's NOT obvious for answering the question of whether the 2 weekend days rained more or less than the 5 other days. Rather it's benefit is in determining which days were the rainiest.

The obvious take-home is to use the appropriate graph to illustrate the conclusion you're trying to make from the data, but also include the raw data so that others can analyse it if they think there might be other details that are important.

wartijn_ 4 years ago | |

I think the "Let’s See if it actually rains more on the weekend." is not really want the author actually wants to test. The text in the image above says "What do you call the day after two days of rain? Monday". This suggests that Saturday and Sunday are the rainiest and second rainiest days of the week.

So what the author wants to see is if both Saturday and Sunday are rainier than any other week. So the problem isn't that you can't differentiate between those two days, but that you can't between either of those days individually and Monday.

marginalia_nu 4 years ago |

The author says the problem is that we have a hard time estimating areas. That is hard, but the problem isn't estimating the area. What you is segments of a circle's circumference. You can unroll a piechart into something like this, and it's still kind of difficult to figure out the proportions:

|----------|--------|---------------------|--|-----------|

asiseeit123 4 years ago |

https://www.perceptualedge.com/articles/visual_business_inte...

- Stephen Few, August 2007

geoduck14 4 years ago |

Yes. Yes. 1000x yes. Pie charts are bad and there is almost no justification for using them. Humans are bad at estimating area - and donut charts help fix this. I actually pretend donut charts are "curved line charts".

Also, engineers are notoriously bad at making visualizations. Sorry if this offends you. I would recommend everyone here spend some time looking at pretty visualizations and maybe reading about it

otabdeveloper4 4 years ago | |

Don't know why you're downvoted; pie charts are exclusively used for obfuscating data for clients in a pretty way.

No sane person who cares about data-driven decisions uses piecharts.

elkos 4 years ago | |

Would you like to share any resources for good practices?

kccqzy 4 years ago | | |

Edward Tufte's various books like The Visual Display of Quantitative Information and Envisioning Information are old but good books.

That said, they are more of an inspiration than a practical handbook of templates to make your own visualization.

mercyandgrace 4 years ago | | |

I really like Stephen Few.

https://www.perceptualedge.com/