« Setting the right priority | Main | Learn EDA (exploratory data analysis) from the experts »



This confirms the conclusions of this study: http://imgur.com/D6EmFhH


This is awesome. Thanks!


And you should link to your full list of uploads, which is great: https://commons.wikimedia.org/w/index.php?title=Special%3AListFiles&user=Junkcharts&ilshowall=1

Andrew Rowley

I occasionally use pie charts, so I was interested to read the links and see why they are considered bad. I agree that it is easy to find bad pie charts, but I'm not so sure that they can/should all be replaced with other types of charts.

Pie charts and bar charts show different views of the information.

Bar charts show values relative to each other. Pie charts show values relative to the total. This is fundamentally different information.

In addition, pie charts, if intelligently ordered allow you to easily see cumulative values, which you have to manually calculate with a bar chart.

For example, looking at the original pie chart, I can see that CFC-11 and CFC-12 are each alone greater than the sum of natural sources. Together, they make up about 1/2 the total. I can work that out from the bar chart, but it is harder to see if you are not looking for it already. It would be more difficult if there were more values in the natural sources.

I think it is dangerous to simply replace pie charts with bar charts, without understanding the information the original author was trying to show.

A lot of the potentially interesting detail in the pie chart is lost in your bar chart.


Andrew: See my previous post about prioritizing your messages. There is no single chart that can provide answers to every possible question. What is the material significance of knowing that CFC-11 is alone greater than the sum of natural sources? I highly doubt that the original author is trying to bring out this particular message. Nor do I think that the two together being half the total is the point of the chart. If that were to be the case, the designer should have color-coded the pieces in a smarter way. Please also read my posts on self-sufficiency. Pie charts are also always a side show because the designer feels it necessary to include the entire data set on the chart. This is a reflection of the deficiency of the chart type, not a strength.


Don't put captions in the image itself. Use the image description page and the thumbnail's caption to describe it. That makes them more easily portable to other languages, reduces duplication of text, leaves more room for the content in the thumbnail, etc.


Andrew, I think a stacked bar chart is better than a pie chart for showing proportions of a total, since we are better at judging line lengths than angles. that's what I tried to do with this https://commons.wikimedia.org/wiki/File:World_energy_usage_width_chart.svg vs https://en.wikipedia.org/wiki/File:Cascading_Pie_charts.png

Andrew Rowley

The section of the article is discussing percentage of man made chlorine, so I would expect this to be the primary focus of the chart. The bar chart relegates the display of this information to the small (pie chart!) legend.

Additionally, the article compares chlorine from HCl to the total from CFCs, and the quantity from CH3Cl to total from man made sources. This information is clear in the pie chart, but difficult to isolate in the bar chart.

Whether the examples I gave are the message the original author was trying to convey is not really relevant. I was just giving an example of information that can be clearly seen in the pie chart but not in the bar chart. However, I suspect the original author would consider it relevant information, in the context of the original article i.e. where does the chlorine in the stratosphere come from, and by implication where to work to reduce it.

Andrew Rowley

"a stacked bar chart is better than a pie chart for showing proportions of a total, since we are better at judging line lengths than angles"

I would agree when comparing lengths of single lines (excluding deliberate optical illusions), but I have my doubts when you are comparing multiple segments within the same bar. Divide a line into segments of 7/12, 1/4, 1/6. Likewise with a pie. Get people to estimate the proportions and see which gets the closer estimates.

A circle gives people natural reference points at 90 and 180 which are very easy to judge - even when a segment begins part way around the circle. Most people would be able to judge 1/4 of a circle even when it doesn't begin at 0. A stacked bar has no natural reference points and I doubt that people would judge 25% in the middle of a bar with any accuracy.


Next step: upload them in SVG format, so it is more simple to translate them in other language, and improve them!



"I think it is dangerous to simply replace pie charts with bar charts, without understanding the information the original author was trying to show."

I think this is a huge over statement. The circumstances in which a pie chart perform better than a bar chart are exceptionally few, and only under an excruciatingly small set of circumstances/purposes.

The arguments against pie charts are extremely strong and well established. The arguments for their continued use are generally flimsy and their refutations strong and well established.

The only time I will ever use a pie chart (other than to illustrate how bad they are), is to show a very small (think sparkline size) comparison between 2 values. A '% on vs % off' type of thing.

The circumstances in which a bar chart is not a better solution are rare in theory, and hard to find in the real world.

The comments to this entry are closed.

Kaiser Fung. Business analytics and data visualization expert. Author and Speaker.
Visit my website. Follow my Twitter. See my articles at Daily Beast, 538, HBR.

See my Youtube and Flickr.

Book Blog

Link to junkcharts

Graphics design by Amanda Lee

The Read

Keep in Touch

follow me on Twitter