# Whatever 25 May 2009

How Many MIT Researchers Does It Take To Screw Up a Visual?

I have no punch line, because this left me speechless. Going way too far in support of a bad metaphor. Wouldn’t some kind of probability plot or histogram have been easier to read? Like this quick sketch by krees.

Speaking of MIT, sort of

From FlowingData I read about a new open-source visualization framework called Axiis, which has been released under an MIT license. I present the visual used to herald this release.

Well, they got some of the the medal counts wrong, but there are other problems.

I’ve reduced their stacked donut wedge structure to a simpler pie chart.

Look okay, until you realize that there were more than eleven countries earning medals. Here is the entire list.

Hmm, about a third ov the medals were unaccounted for, and there’s no indication of that in the chart. In this bar chart I’ve lumped the missing 1/3 of the medals into an “Other” category.

Another problem with the Axiis chart is shared by all pie-type charts. The points have no common reference point: the starting points of each angular measure starts where the previous measure ends. Can you tell from the Axiis chart whether the US won more gold, silver, or bronze medals? Did China or Russia win more silvers?

Plot the data in a boring old bar chart, and ask the same questions.

Much easier to make comparisons.

The software itself looks interesting, but judging from this and other samples, there seems to be no guidance to help the user avoid ineffective chart designs.

Bill Gates Does TED

Bill Gates poses a couple of problems for us to solve. How do we eradicate malaria, which is one of the three largest contributors to childhood deaths, and how to we transfer good teaching practices from the top 10% of teachers to the rest, in order to bring the US back to parity with thee best in the world.

1. Matthew says:

Your charts are much better, but I don’t think the three bars in one big bar reallly works, as the area is 1/3 the size it should be and so it’s hard at first glance to understand the grey bar in relation to the other bars. What could be done is put them consecutively the same width as the grey bar, the problem with this is like any stacked bars you can’t compare the size of the silver/bronze as they don’t start at the same point.

2. Matthew -

I wanted to get away from the stacked bars, which means clustering the medal types side by side. Clustering the total next to the other bars doesn’t work, either, as it gives the total the same apparent weighting as the separate medal types. The question then is how do you show the medal totals as well as the individual medal types.

You can use an outline instead of a filled bar:

and even use a dashed line to further distinguish the totals

I don’t really like the dashed outline, but the solid outline isn’t too bad. Alternatively the totals could be shown in dot plot style.

With or without connecting lines.

At some point you have to rely on the intelligence of the user to figure it out. As charts go, this one isn’t too tricky.

3. Actually, I kind of like the three in one bar approach. I was able to pretty quickly work out the individual counts as well as how in compared to the totals. I’d lean towards the white background with solid line for the big bar though.

4. Thom Mitchell says:

In contrast to Matthew, I really like the three-bars-in-one-big-bar charts. Further, my preference is for the original, gray-shaded total lying underneath the three “disaggregates.” I think it effectively conveys information on both levels without adding a separate chart, even if purists would prefer or demand exactly that: an additional chart.

5. Tim says:

I think the MIT warming chart is ok.

I looked at it for a moment, and I could tell that they predict warmer temperatures with no policy and in that case we are most likely to get to somewhere in the range of 4-6 degrees increase.

I think a caption that said “probability of increase in temperature” would have made it even quicker for me, but I had no real trouble interpreting this chart.

6. Peter Jung says:

In defense of my alma mater, I don’t think their “chart” was purely intended to be a data visualization. If you look at the photo in the article, you can see that it’s a prop, meant to be spun like a wheel of fortune. Sure, the design fails if you’re purely interested in the numbers, but I think the roulette wheel metaphor, reinforced by the “greenhouse gamble” label, isn’t really all that bad. OK, maybe it’s a little cheesy…

7. Yeah, it’s my alma mater as well (III ’81). And as often happens with clever charting metaphors, it fell flat. See Spiraling Down the Drain for an example of another failed metaphor.

8. @Jon, qualifier: I am the guy who started Axiis and one of the founders :) I am also the person who designed that wedge chart (which has actually been a design others had conceived of before me.) Great analysis and I agree with most of your points. I personally loathe radial charts for the very reasons you enumerate so well with your examples, but i did just want to point out a few details.

1. That wedge chart stemmed from a clients request where the data MUST be seen in a pie format, full discourse can be seen here: http://www.twgonzalez.com/blog/?p=143

2. Axiis does ship with some prebuilt visualizations, but they are simply the starting point. The purpose of Axiis was not to create the standard charts you see everywhere, but provide a platform that allows you to make your own visualizations based on your own needs. I encourage you to take a look and apply some of you great visualizations skills with the framework.

Regardless, good critique.

Best Regards,

- Tom

9. Tom -

Thanks for your comment. I understand the pressures that clients can bring to bear. I’ve also walked away from several projects that involved ill-conceived graphical constraints.

A good craftsman doesn’t blame the tool, but you should be careful that Axiis does not become associated with the set of tools that those who like shiny objects call BI. A few too many glittery templates or examples might put it in the class of a Dundas or a DotNetCharts. Excel itself has severe problems with poorly designed defaults and templates, and Excel also makes it hard to overcome these defaults to produce effective visualizations. However, it’s just a tool, which can be used with good results. I imagine Axiis can as well.

By the way, in Excel that type of chart is called a donut chart. It may not be as fancy, but in effect it’s the same style.

10. @Jon, yep it is a fine balancing act to put enough shiny objects out there to attract attention and get an audience, balanced with good solid data viz. For that one specific project the wedge graph represented less than 1% of the work, so didn’t want to throw out the baby-with-the-bathwater so to speak.

I personally believe aesthetics don’t have to be the antagonist of good data viz communcation, and that just because you have a “non-essential” aesthetic flourish does not mean you have corrupted the communicative properties of the data visualization. If you look in the Axiis sample gallery we actually do have some charts very similar to what you have shown with the same (incorrect) data as the wedge graph: http://www.axiis.org/examples/HClusterStackExample.html

Now that we have a release with some shiny objects I hope to post more examples of more significant visualizations that explore some more advanced techniques both technically and from a data viz perspective.

11. Perhaps some sparkle and shine is necessary. But once you have someone’s attention, it’s important not to put form over function. Aesthetics doesn’t have to be the enemy of effective communication, but often what some call aesthetics is the creative abuse of colors and effects. While I think Tufte occasionally takes it to its illogical conclusion, his brand of minimalism greatly enhances information transfer. Keep It Simple Stylistically, or something.

12. Thom Mitchell says:

I still like Jon’s first two three-bars-in-front-of-one-bar examples (with the gray shading). My reasoning remains the same: I think it effectively conveys information on both the aggregated and disaggregated levels without adding a separate chart. If purists would prefer or demand two separate charts, I won’t argue against that, but I think the Axiis example of a “cluster stack” is ill-advised for two reasons.
(1) setting the stack next to the three individuals implicitly suggests to me a fourth, independent datum, which it is not in this case;
(2) giving the stack three colors and including the values is redundant and crosses the “less is more” line.
For what it’s worth.

13. @Thom, good point, and I would probably agree with you. We would love to have people like yourself, Jon and others contribute examples of best practices like the above. My goal in designing Axiis was not about pretty/shiny as much as expressive and making it trivial to do what you want in a visualization, without feeling like you are performing a bunch of bubble-gum and tape witchery in applications like Excel, or with 3rd party dev SDKS. The examples are meant as more of a showcase of the flexibility of framework code (visible by right clicking for view source) and less about best practices. Being a completely volunteer effort we were scrambling to meet a deadline we set. I am hoping over time to have more compelling “best-practice” examples. Thinking about implementing the famous Minard – Napoleon March diagram.

With more people in the community to contribute examples I would love to set up a “best practices” category on the website where visualizations could be voted on by the community against different criteria like “communication effectiveness/integrity, aesthetic appeal, ease of understanding, etc..”

14. “expressive and making it trivial to do what you want in a visualization”

A lofty goal, compared to my duct tape witchery ;-) and the learning curve of a capable programmatic approach like R. I already appreciate the flexibility demonstrated in the examples on the Axiis web site. Is there a user interface wrapped around the XML, or do you still generate code manually?

15. @Jon, no I made a conscious effort to not use a User Interface, (after building one for 2+ years at a startup I co-founded.) My experience in building WYSIWYG tooling for dashboards and data-viz, and using it, was that no matter how flexible you made it you ended up restricting the freedom of expressiveness because of lowest common denominator (metaphorically) scenarios, or you ended up building such complex tooling that would have a higher learning curve than just programming. Axiis can be created in something as simple as Notepad and run against the compiler, or something a bit more integrated like the Flex/Ecplise IDE

Axiis is designed to leverage a markup syntax (versus more complex OO/procedural programming, which you can use as well) and thus be approachable by a wider audience. But it does require you to “program.” After getting the hang of it, I think people will find it to be a liberating experience, and much more expressive and free than using pre-built tooling. I am thinking about providing an web based markup editor that allows online compiling so people not familar with Adobe Flex (or who just want to play around) can do so online, build some visualizations and have and portable output (.swf file) as the result.

16. Tom -

I didn’t really have a WYSIWYG kind of interface; the complexity of such an interface is daunting. I was imagining some kind of form-driven interface that would provide options to users, and would take their input and generate the markup.

The problem is that while this markup approach lends much flexibility to a programmer, it is still “programming” to a non-programmer. The need to learn th arkup languae might be a significant barrier to entry for those who would consider using a simple interface to build the code behind the scenes.

It’s not conceptually different than an Excel add-in I might write to collect user input through a series of dialogs and build an Excel chart using Excel’s VB object library. It provides less flexibility than allowing a user to write their own code, but it eliminates their need to do so.

17. As an example, I’ve been intending to spend a little time learning R. I downloaded it months ago, and I’ve collected links to some decent examples. I’ve seen what it can do, and it’s impressive. I’m not afraid to learn a little programming, but for me now the barrier to entry is simply the time it would take to become the least bit proficient in R. The Axiis markup language presents this same barrier for me, a semi-competent programmer.

For a non-programmer, both R and Axiis are non-starters. When I was still in “Corporate America”, I might have been able to develop skills in such a platform, but I could never have transferred this capability to my colleagues or superiors. Now as an independent, I can still consider learning these tools, but again, they are not readily transferable to 99% of my clients. If I could hook into an interface, that would change the equation.

The problem is that a lot of packages that have interactive interfaces produce such, excuse me, crap. But because they have the interface, they are usable (or abusable) by a broad spectrum of users. Off the top of my head, I can’t think of a package with a friendly interface that produces effective visuals.

18. @Jon, and therin lies the rub. I agree with you, learning markup or some type of programming will create a barrier of entry that excludes a lot of people. And that exercise of “bringing visualization to the masses” is probably a loftier goal than I am am personally ready to pursue at the moment :)

Your form based idea sounds intriguing though.

Personally I see no reason why someone who can do what you do, and has a true vested interest in producing outstanding data visualization wouldn’t be willing to spend a couple of weeks to become proficient in a tooling or language that allows them to explore new levels of expressiveness and communication with data be that in R, Axiis, or one of the many other platforms available.

19. Tom – It’s my stated intention to spend some time to expand beyond Excel’s charting engine to something like R or perhaps Axiis. What prevents me so far is the limit of 24 hours in a day, and I have to waste six of them sleeping.

20. Thom Mitchell says:

Some time ago I used MathCad (by MathSoft) for its “symbolic math” capabilities. MathCad 2000 came with a data visualization app, Axum 6. I installed it and gave it a look-see, but Axum certainly didn’t “take” with me.

I use Excel for the computation and the relative ease of programming. Its ubiquity doesn’t hurt, either. (My daughter works in sales and that’s how she keeps track of her contacts and her performance.)

When it comes to charting — “presenting”, “representing” or “visualizing” the data — I bump into two barriers. (1) Excel isn’t perfect and (2) my knowledge of what Excel can do is woefully incomplete. If I’m certain how I want to present the information and I can’t figure out how to twist-and-tweak-and-coax Excel to do it, then I resort to Google. If Google doesn’t find me a solution, then I politely beg my favorite Microsoft Excel MVP. :-)

The clincher for Excel, though, at least for me, is the connection of the chart back to the underlying data: change the data and the charts update.

Disclaimer (possibly of interest to Tom G). As an economics professor, my own data visualization needs are relatively straightforward: I am generally plotting curves in 2-space; frequently, these curves are step-functions. In some of my graphs, I like to have the region under a curve shaded. That’s about it for my visualization needs, so I am unlikely to commit the time to learn Axiis (just like Axum 6 several years ago).

The next “level” for me is to exploit Excel’s connection between data and charts to make the charts interactive and respond to user (i.e., student) entries, whether the entries are typed in cells or generated by spin-button clicks.

Recently, however, I have refereed journal submissions of empirical papers. My discipline is horrible on interpreting empirical results. There are generally paragraphs upon boring paragraphs devoted to highlighting a selection of the numerical results in one or more tables. The referee report I submitted last week encouraged the authors to use a simple column or [horizontal] bar chart to highlight their number-crunching results. I am probably a sandbag-against-the-tide in my request for “results visualization,” but I can hope. This didn’t figure in my referee report, but encoding two types of information would be straightforward: let the bar height or length represent the value of the estimated coefficient (positive or negative) and use color (or merely dark and light shades of gray) to encode the presence or absence of statistical significance.

21. quick update. I finally found a couple minutes to get back to this, and with about 3 lines of code change, re-graphed the chart to something more along the lines of Jon’s excellent example.

You can see it here: http://www.axiis.org/examples/HClusterColumnExample.html

22. Thanks Tom. Looks pretty good. If that only took a few lines of code to change, you’ve got a flexible framework.

23. well, it could be dialed into be a more “pure” data viz and really maximize the chart-ink –> info ratio. but yes, it was probably about 50 characters of code to change across three XML attribute properties.

24. I don’t mind a little chart ink used to make a chart attractive, as long as you don’t gussy it all up.

25. Martín says:

Jon,

as dumb as I may seem, I have to ask: how did you get to make that chart? I need to do something similar, and clearly have no idea on how to do so.

Would it be too much to ask you to share the file, or briefly tell me how to create the chart? I am using Excel 2003, with no add-ons.

Thanks !!

26. Martin -

You mean the Olympic Medal bar chart? The total series in on the primary axis and its gap width is 50%, the individual medals on the secondary, I’ve added the secondary category axis, and I’ve removed the secondary value axis.

27. Martín says:

Thanks Jon ! actually, by the time i got your reply, my brain already started up, and figured that out. quite impressive chart, and really, really simple !

A good story to tell…

28. Martín says:

Jon,

I’ve just saw Bill Gates’ video. Impressive approach, and quite simple.No wonder why he’s such a visionaire…

Subscribe without commenting