Almost never use a pie chart for data | Hacker News Reader

Almost never use a pie chart for data

theconversation.com

41 pointsBrajeshwar24 comments

I'm glad they explain that beyond 2-3 data points is the problem here. I'm working on an interface that will us a lot of pie charts, but they are all representing 2-3 distinct things.

I'd argue even 4-5 things can still be ok given they're significantly (enough) different sizes. Recently I needed to show how our online revenue was split between 5 different products and the pie chart worked perfectly.

With favorable numbers pie charts can be readable, but what makes them better than a bar chart?

At least to my brain, showing 'parts of a whole' is easier to intuit in a pie chart than a bar chart. Again, the caveat being there aren't more than 4-5 parts and the sizes are different enough.

That said, I'm now interested in asking around the office to see what others think. Maybe I'm an outlier and should switch to bar charts.

I would still argue that a stacked horizontal bar with tick marks or a waterfall chart are often the better solutions. They also make it easier to provide a reference point for the absolute sum of all data points

I think pie charts only work in those cases where it doesn’t really matter how big individual data points are, but where one only wants to give a feel for the relative weight of 2 (max 3) data points. But then: wouldn’t a simple KPI (“55%”) suffice?

Stacked bar charts have the same weakness as pie charts where it is very difficult to tell the relative sizes of the pieces.

Edward Tufte is highly recommended on the topic.

It's not a hard rule, more encouragement to investigate ways to make the presentation more informative by relating the data to it's context, time etc.

A pie chart is one-dimensional, quantities period.

His book "Envisioning Information" was very influential for me.

Generally agreed. I think 4-5 data points is fine, but the advantage of pie charts is that it's a quick visual representation of what proportion of the 100% can be attributed to the 4-5 segments.

Pie charts fail again once the distribution is very skewed (1 segment = 99% and the other 3 shared 1%). It also fails when the distribution is too even (5 segments 20% each). They shine the most when there is variance, but overall easy to interpret the distribution differences.

Pie charts shine when there several data points that can be grouped into 2-3 macro groups.

The easiest example is "What is your opinion of ___?" (Strongly Favorable, Somewhat Favorable, Neutral, Somewhat Unfavorable, or Strongly Unfavorable)

At a glance, you can see if the opinions are positive or negative, and then can see if they are polarizing or mostly neutral.

Another example would be the makeup of congress. Democrat, Independent, and Republican for the major groups. Subgroups could be based on percentage of votes the person received from their district. It would show which seats could swing.

There is an advantage of using pie charts when you have lots of categories, especially when some are tiny slivers. It's that you can see which items are negligible in terms of their relative contributions. A world GDP pie chart gives a good representation of how international GDP is distributed, and which countries represent only a sliver of the total.

EDIT: Typo

Wouldn't bar charts show that too but without the issues mentioned in the article?

A bar chart is better at showing each individual value independently (and perhaps in relation to its neighbors) but is worse at showing each value as a proportion of an overall aggregation.

Yes, you can visually add slivers. Just ignore the boundaries between them, and boom you've added them.

If you found this interesting and you do any data visualization work in your life, I can highly recommend Information Dashboard Design by Stephen Few and The Visual Display of Quantitative Information by Edward Tufte. While some of the examples certainly appear dated these days, the concepts, best practices, and approaches to data visualization outlined in these books, such as the concept of data-ink and non data-ink, are timeless in my opinion. I actually can't remember how many times I've bought these books as I now work as an engineering manager and have the habit of giving away my copies to new developers.

> While statisticians have criticised them for decades, it’s hard to argue with this logic: “if pie charts are so bad, why are there so many of them?”

A bar chart can represent many things. When you see a bar chart, you have to ask whether it's a trend over time, portions of a whole, values meaningful relative to each other (like benchmark results), etc.

When you see a pie chart, you don't have to ask. It's portions of a whole.

Pie charts communicate "what am I looking at?" instantly.

I actually prefer pie charts when information is similar and you're trying to show relative sizes.

The example in the article shows why: in the example of the three, it's asking "which segment is biggest?" But that's the point, they are all similar. You don't need to know which one is biggest, or: only if you do need to know that, would it be the wrong way to display information.

I'm currently making a budgeting app for my friend, and I'm using pie chart for expenses. We don't need to know "which is the biggest"

It's more visually clear that there's broad categories for expenses, and how they sit "relative to the whole" - drawing it as a bar chart would be misleading, because it doesn't show "X is roughly the same size of Q,R,S put together." Without the observer doing some maths and mental "stacking"

seems like the bar chart examples (as a solution to the bad pie charts) isn't representative.

if you're talking about a pie chart that is fractal of smaller and smaller wedges, it seems like you aren't going to achieve much by going to bars.

you probably just wanna do a grouping of all the smaller ones; other.

my take away is, don't have too many, 5 is the aesthetic max, where 2 or 3 is ideal.

personally, stacked bars are only useful if you're showing a trend. otherwise, its a question of density, not format IMHO.

I think pie charts are fine provided three things:

1. The pie chart should be organized so that starting from the top and going clockwise, you go from largest to smallest data point. The example chart of COVID data in the article does this correctly.

2. No more than 10 data points. Beyond that, it gets too cluttered and less meaningful.

3. DO NOT MAKE THEM 3D. Making a 3D pie chart (Or really, ANY 3D chart) should be a crime. 3D charts (that have no reason to be 3D) sacrifice a chart's ability to show data, the entire purpose of a chart, in favor of silly aesthetics.

My prrsonal pet peeve:

don't use a pie chart for a [free/used] disk space/RAM.

I'm looking at you, Zabbix.

Especially for %used, it's meaningless.

My favorite pie chart: https://flowingdata.com/2008/09/19/pie-i-have-eaten-and-pie-...

My second favorite: https://xkcd.com/688/

Besides being informative, the article is also well written, I learned from it easy-to-remember bits of info.

Not to mention the fun style of writing:

"Possibly they are popular because they are popular, which is a circular argument that suits a pie chart."

I'm glad that doughnut charts came to rescue us from our pie chart hell.

Donut charts are atleast fun compromise between stakeholders wanting pretty charts with less misleading flaws.

They're stacked bar charts in the polar coordiate space, after all.

Reply on news.ycombinator.com