There are some things that I have learned over the years as bad practice. There are some that I have learned through reading and contemplation, and a little experimentation. One of the categories of bad things that I have learned is visual aids for statistics. There is basically one cardinal rule for data presented in a graphical form:
Let the user see the actual data.
It seems kindof obvious, doesn’t it? I’m presenting data, you should be able to see it. If you want to show me that there are 450 widgets, and 230 whazits, and only 210 whositz, then you should show those. The simplest way of doing this is by just listing them out in a table:
| Type | Amount |
|---|---|
| Widgets | 450 |
| Whazits | 230 |
| Whositz | 210 |
That’s simple, isn’t it? However, it doesn’t help the viewer any. They’re pretty good if you want to know exactly how many of something there is, but not great at other things that you want to do with data. How do you find the minimum (or maximum) of all of the values? You scan, remembering the highest one you saw so far, and then go back to that one and find out. If the goal is to somehow compare, the viewer will be helped out by using a graphical display of some sort. One thing that humans are good at estimating is length. Quick, tell me what we have the least of:
That was easy, wasn’t it. Bar charts are good - they can still be abused by people in evil ways, but most of the time they’re pretty useful.
All of this is working up to pie charts. What’s smaller, Whazits or Whositz here:
Is Widgets half or more than half of the pie? Pie charts are the bad seed of the graph world. They aren’t very useful, hang out a lot, and don’t help you much. The worst thing about pie charts is that they aren’t even good at the thing they’re supposed to be the best at: comparing relative sizes. Consider thanksgiving, or christmas, or whatever large gathering you prefer where there is pie. Lots of pie slices are laid out on the table. Which one is the biggest? It’s hard to tell. It’s even hard to cut even slices from a normal pie. Sure, you can cut in 4ths, 6ths, or 8ths, but I challenge you to cut a pie in 5 pieces equally - it’s really hard! It’s because humans aren’t great at judging the differences between angles. They’re only really good for seeing one thing: the largest piece in the pie. Even then, it’s sometimes questionable - if you looked at a evenly sliced pie, could you tell the difference between a 23% piece from a 20% piece? The bad thing about this is that they are ingrained into our culture. There wasn’t a finance report in the last 10 years without a pie chart.
All of this ranting was basically brought to you by a recent post on BoingBoing. This post features the most horrible monstrosity
I have seen in a long time: the hierarchial pie chart. Feast your eyes on this monstrosity. I challenge any reader to tell me: what is the smallest third-level category? What is the largest?
Comments 13
Why that’s so obvious! The smallest third level category is Business as overlaps Science in which there is a favorable mention of DRM and the largest is naturally the Arts ( which means anything about Cory, Cory’s books, or Cory’s lecture series!!!)
Posted 08 Jan 2007 at 10:11 pm ¶It is a pity that you picked one of the few examples where pie charts are not _that_ bad, namely when the categories are close to 25, 50 or 75% of the total !
Indeed, you ask “Is Widgets half or more than half of the pie”, and your implicit answer is that pie charts are not good for this. Well, in the present example, bar chart are much worse at answering the question: there is no way to get this information from the bar chart, while I would argue that we can see on the pie chart that “widgets” are >50% of the total. Yes, it may be hard to see because if is very close to the total, but it is still easier than on a bar chart (noone can mentally add the heights of Whazits and Whositz and see that the sum is just below Widgets).
As you say, humans are not good at judging angles, *except* for right angles.
Posted 09 Jan 2007 at 1:16 am ¶There is a place where a similar monstrosity IS userful:
http://www.methylblue.com/filelight/
Posted 09 Jan 2007 at 2:57 am ¶Yes, in KDE, there is a view of disk space which shows the file system in a very similar monstrosity - it is quite helpful.
Posted 09 Jan 2007 at 5:45 am ¶I find that that is pretty bad still. I prefer a different view:
http://www.derlien.com/
It has the advantages of being hierarchial just like the expanded pie chart while also being quick to see the largest files and directories, which is what I’m looking for most often.
Posted 09 Jan 2007 at 10:22 am ¶This is very true - the example isn’t that bad for a pie chart actually, and anyone who looked at it would be able to discern that one is about half, and the other two are about 25%. I should have chosen the data more carefully in order to illustrate my point.
Posted 09 Jan 2007 at 10:32 am ¶Just imagine what the Lie Factor would be on those pie charts…
There is a reason to use pie charts (although not very frequent) and yet there are people who still use them exclusively.
So the only conclusion that I would draw from those charts, is that the author did not want me to understand the data.
Posted 09 Jan 2007 at 11:21 am ¶We can note, however, that the bar chart could be improved by adding thin horizontal lines showing 25% and 50% of the data; this way, we would also be able to see that the data is split in 50%, 25% and 25%, thus making the bar chart superior to the pie chart (as it is normally).
Conversely, the pie chart would usually display the number of observations in each of the three slices (the equivalent of the scale in the bar chart), so that the raw data would still be displayed.
The Wikipedia entry at http://en.wikipedia.org/wiki/Pie_chart has some good references on the topic.
Posted 10 Jan 2007 at 12:09 am ¶“The worst thing about pie charts is that they aren’t even good at the thing they’re supposed to be the best at: comparing relative sizes.”
If similarly sized slices are the main problem, just sort them (which is not obvious in your 3-slice chart, for which only one sort exists).
Posted 14 Jan 2007 at 12:15 pm ¶If you need to rely on a cultural constraint in order for your data visualization to work properly, I suggest that you should use a different visualization method which doesn’t rely on implied information.
Posted 14 Jan 2007 at 12:23 pm ¶Surely the ideal would be to use some sort of Web 2.0 app (and there must *be* one, just not one I know) that contains the data - so the user can switch between raw data, bar & pie chart, depending on the user’s preference. (E.g. - if you have an SVG viewer - http://www.adobe.com/svg/demos/chart.html )
(It would also allow the user to select correct colours. To me, the biggest problem by far with the quoted pie chart was not that it was a pie chart, but that it was in shades of red & green … )
Posted 15 Jan 2007 at 8:36 am ¶Indeed, the ideal is probably some interactive way of going through the data. The principle of showing the data still holds though.
I do try to pick colors which are distinguishable to all viewers - the colors look magenta, blue, and yellow to me.
Posted 15 Jan 2007 at 12:11 pm ¶Hi Michael. It was interesting to see your reaction to my multi-level Pie Chart you saw on Boing Boing. I’ve written a response that you and your readers might find interesting at:
Posted 10 Feb 2007 at 7:35 am ¶http://www.neoformix.com/2007/InDefenseOfPieCharts.html
Trackbacks & Pingbacks 1
[...] Learned Bad Ideas, by Jamuraa, Base Zero [...]
Post a Comment