Data Viz Done Right

June 8, 2012

Donut charts are worse than pie charts: Have no fear! There are many alternatives.

16 comments

Ironically enough, donut charts are basically pie charts with a hole in the middle, leaving you with even less information to fill your brain.

Donut charts are used to display data in rings, where each ring represents a data series.  Each ring is a distinct category.  Yes, there are examples of donuts within donuts (here is a particularly horrific example from Dundas). 

Once again, Nielsen provides another example to pick on review.  At least their consistent in their ineffectiveness, right?

Pie charts can be effective, rarely, but I don’t see why you would ever use a donut chart over a pie chart.  Let’s look at some alternatives to Nielsen’s infographic.

Alternative 1 – Pie Chart

image

Grade = F

  1. It’s a pie chart
  2. It’s a pie chart
  3. It’s a pie chart

Alternative 2 – Bar chart by Device by Category

image

Grade = C

  1. Comparisons between Devices is difficult
  2. Data is sorted by frequency of using the Device, making it harder to see the winners and losers
  3. The coloring by Category is distracting

Alternative 3 – Bar chart by Category by Device

image

Grade = B

  1. Easy to compare devices, which was the point of the article
  2. Labels removed since the exact values are not all that important and distract from comparing the size of the bars
  3. Devices are sorted by frequency within each Category, making it easy to see the winners and losers in each Category
  4. The colors neatly segment the data

Alternative 4 – Heat Map

image

Grade = A

  1. Easy to compare all of the data at once, whether you want to look vertically, horizontally or both. 
  2. A heat map makes it clear that people use their computers at home to watch video most frequently and that people watch a LOT of video, independent of the device.
  3. The colors are impactful and provide rich meaning, while not overwhelming your eyes.

Alternative 5 – Highlight Table

image

Grade = A+

  1. You get all of the benefits of the heat map, but you also get the values, providing additional detail.

These are only five of the many possible alternatives.  Download the Tableau workbook here.

Would you use any of these charts to visualize this data?  Why or why not?  What other alternatives would you find effective?

16 comments :

  1. Great post as always. I definitely like Alternative 3 because I feel I can better compare the values between each category with a bar chart. However, the heat map is very effective for getting a quick understanding of the overall story.

    The one thing I would add to the 3rd alternative are vertical gridlines. The chart is tall and I think the gridlines would allow an individual to better compare the values without having to keep glancing back and forth between the bars and the axis at the bottom.

    Nice work!

    ReplyDelete
  2. Robert, I intentionally left the gridlines off because I didn't want the viewer to get caught up in the precision of the values, but rather the shape of the data.

    ReplyDelete
  3. Andy :)

    I have been waiting all my life to see it "Alternative 1 – PIE Chart" :)
    I thought that you would never write it :) Thank you :) so there is something what you hate more than a pie chart...

    ReplyDelete
  4. Yes, donut charts are worse than pie charts! And radial charts.

    ReplyDelete
  5. Nielsen is giving you a lot of work

    ReplyDelete
  6. Nice post. I'm with Robert. I like alternative 3 as well as 5. They emphasize different aspects of the data so I like showing both.

    Russian Sphinx - He's not alone. I'm in the middle of writing a blog post where I claim that something is worse than a pie chart. Never thought I'd be saying that.

    ReplyDelete
  7. You did not offer a dot plot as an alternative. You could plot with the same dimensions as either bar chart, and it would be at least as good as the bars.

    I prefer the third option and the missing sixth option to either heat map. Shaded fills aren't as effective at showing patterns as bar lengths or dot positions.

    ReplyDelete
  8. I'm glad I didn't see spider charts. They are nearly as bad as an array of pies or donuts.

    ReplyDelete
  9. Jon, I've created the dot plot and agree that it's effective.

    http://farm8.staticflickr.com/7090/7354341226_ecac4f7ca4_o.png

    I also created a lollipop chart, which looks pretty cute, but isn't as effective as the dot plot.

    http://farm8.staticflickr.com/7075/7354342090_fb3fe0f4da_o.png

    I agree that heat maps aren't as effective for showing patterns, but I like them for fast comparisons.

    ReplyDelete
  10. Andy, I think #3 (with a few improvements) does a better job than the heat map. Here's an edited workbook that contains a few changes:

    http://public.tableausoftware.com/views/NielsenGlobalReport-Multi-ScreenMediaUsage-SW/Bar2-SW

    Specifically I think you're wasting color using it on category, instead of device. Also I think the sort doesn't work that well since there are only 3 devices per category. I think it's more effective if each device maintains a consistent position (and color). I also added a filter for device so it's easy to pull them out of the pack.

    Of course these are all minor changes, and you're right any of your options beats the original donuts hands down.

    --Shawn

    ReplyDelete
  11. Well done Shawn! I really like your version of alternative 3 better than mine.

    ReplyDelete
  12. Andy, I actually believe that we can use pie charts from time to time, depending on the task at hand and the audience. Never had the opportunity to use them myself, but I'm patiently waiting.

    I like donuts. Remove the hole and you get a nice multilevel pie (treemaps are unavailable for us, poor Excel users). But it's risky to use them the way Nielsen does (comparing round charts is a capital sin, so you'll burn in hell). This applies to alternative 1 also.

    Alternatives 2 and 3 should get the same grade: they are two different perspectives of the same data, using the same chart. And I don't really like bar charts. They are overrated.

    Heat maps are nice, but the moment you start using them you must stop complaining about how difficult is to compare angles in a pie chart :).

    My own preference goes to connected dot plots, or profile charts like this: http://ow.ly/i/RfQi/original.

    ReplyDelete
  13. Jorge, I challenge you to show me a single example of an effective donut chart. It's simply impossible to compare the sizes of the slices. I can't see how you would ever prefer a donut or pie over a bar chart and feel as though you've provided the best available visualization.

    As for your design, you've provided a perfect example of how you should NEVER connect categorical data with lines. Lines imply trends. You're giving your reader the impression that there's a trend to be found, when that's simply not the case with categorical data. Overall, it's a poor design decision in my opinion.

    Thanks for the comments. These are great discussions.

    ReplyDelete
  14. Jorge, you might also find the explanation/discussion on pages 132-135 of Few's Information Dashboard Design useful when looking at your version.

    Your's would work well as a dot plot.

    ReplyDelete
  15. People are unable to judge the relative area of circles and this makes many visualization types misleading: dot map (like one in background on this blog), bubble chart. Unfortunately, this is where the designers and the data nerds part ways, because circles are better looking than all the linear and rectilinear options discussed above. Doesn't the Nielsen example look better than all the other alternatives (forget for a moment that it is unusable)? So the trick is sometimes to make a good looking visualization that is also not compromised too much in usability. This takes finesse. Does a pie chart suffer from the circle area problem really? One should read a pie chart by ignoring the center and looking at the relative arc lengths around the perimeter. Many claim this is easy for them. I think there may be a problem in that one will discount the length of curved lines more and also that lines next to each other will affect the perception of each other's sizes. Of course this is a problem in many other visualization types. Does a donut chart suffer from the circle area problem? Well it depends on the difference between the inner and outer radius I think. If they are too different, it approaches a pie chart and has the same issues; if they are too similar, it is difficult to compare the relative lengths because the eye has to move around across that distracting negative space (e.g. Nielsen donuts above). I think smaller pie and donut charts are easier to read also for this reason. This is part of why Alternative 1 above works better than the Nielsen example. The worst thing about the Nielsen donuts is the color scheme chosen totally obscures the data. If they had used a one-direction color ramp, this would have been largely alleviated. Here is my example: http://maps.rlid.org/Misc/ChartTypes.htm. Note the stacked bar chart is very similar to the donuts (relative baseline issue), but is more compact.

    ReplyDelete
    Replies
    1. This makes most sense to me. It is always a challenge to get the data smart folks to agree with UI specialists. UI folks make the charts look cool, Data folks want it to be useful. The key is to get them to agree on a via media that achieves both goals to the maximum extent.

      Delete