<-- Advertise here.

Napoleon's March.jpg
[infosthetics@strataconf 2011 by guest blogger Collin Sullivan]
The above graphic is 150 years old, and it popped up more than a few times on Wednesday, which was Day 2 of the 2011 Strata Conference. Charles Joseph Minard designed it in 1861, and it depicts Napoleon's ill-fated march to and from Russia nearly 50 years before. The line is light as Napoleon and his army march toward Moscow, and solid black as they retreat. The line's thickness depicts the size of Napoleon's army at the corresponding place on the map, and we can follow along as it thins dramatically over time. [see it also here]

Minard's graphic is a classic in the field, as it should be. It is considered both beautiful and parsimonious; from just a small amount of space we can glean a lot of information. Perhaps the single characteristic that makes this map most effective, though, is that it tells a story.

Pete Warden of OpenHeatMap believes that humans are natural storytellers. In the Visualizing Shared, Distributed Data panel discussion on Wednesday, he told us, "Our brains are hard-wired for stories." Stories are how we communicate with one another, how we share information. Narratives provide the context within which we make sense of data. And in talk after talk, data visualization was presented as a medium by which we can tell a story.

It makes sense. The information packed so neatly into Minard's map would fill dozens of spreadsheet rows, and writing the information out in narrative text form would be laborious and, in the end, likely ineffective. Data simply lends itself to visualization.

But as is true with all types of storytelling, some approaches are more effective than others. That was the topic of two of the morning discussions Wednesday, one given by Tableau Software's Jock Mackinlay and the other from Kim Rees of Periscopic. Rees spoke on visual economy (Small is the New Big: Lessons in Visual Economy, slides available here), focusing on the succinct and the necessary. Graphs and charts can and ought to be beautiful, but it can and should be done without extraneous information and without wasting space. She showed a beautiful infographic displaying attempted missions to Mars as an example:

Missions to Mars (viz economy).jpg
The graphics are close but not cluttered, and the information is plentiful but not overwhelming. The chart is concise and complete; in a word, succinct. Rees emphasized that word, among others that suggested things should be small, compact, dense-but-readable. Most importantly, though, the narrative is clear. We can see successes and failures, visually distinguish between countries, follow along on a timeline. Another great example she used:

Flight Controls 2.jpg
This tells a pilot all she needs to know about the relationship between the plane and the horizon. Including additional numbers or degree measurements would add only clutter here, as this quickly answers the only question the pilot is asking at the moment: Is the plane level? There is no need for trees or further landscape design, no labels for the sky or the ground. The color selection does that work for us. The context is minimal but the information is adequate.

If Rees' talk was about how to visually frame the narrative, Mackinlay's presentation was about how to tell the story itself (Telling Great Data Stories Online). He implored us to allow the data to decide the visual medium. That is, we should not find a graph we like and force the data to fit into it. Instead we should identify which graph fits best for this or that particular data set. The visualization, he said, ought to communicate all of the relevant information, but in a way that leverages what Mackinlay called the "human perceptual system" (remember the Gestalt Laws from Tuesday afternoon's talk with Naomi Robbins). Here is an example he used:

Tableau Bad Good.jpg
The graph on the right works better because it uses color, rather than size, to identify the variables. Our brains automatically assume that if a data point has a greater area, it is somehow larger than or superior to the other data, and it draws our eye. But in the graph on the left, size is used only to suggest differing regions, the area of which is not important here. It communicates something that it should not. (Robbins bemoaned bubble charts partially for this reason, and also for the difficulty of comparing relative sizes in a small space when the differences are subtle).

The best things a person can do to become a better storyteller, Mackinlay told us, is first to look at good examples of visual storytelling, and then to have a proofreader or editor. The presentations at this conference have provided a wealth of the former, and there are plenty of places on the web to find more (Tableau, Google Fusion Tables, The Guardian's DataBlog, Periscopic, etc.). And as storytelling emerged as one theme of Day 2, the question of proofreader or editor segues nicely into another prominent theme, that of dialogue between user and graph. My next dispatch from Strata will explore some of the ways this idea manifested on Wednesday.

This post was written by Collin Sullivan. He is a research analyst for The Sentinel Project for Genocide Prevention, where data collection, analysis and visualization are being used to design an Early Warning System (EWS) to detect and prevent genocide. Collin lives in San Francisco. You can reach him at collin [at] thesentinelproject [dot] org and follow him on Twitter at @inciteinsight.


Regarding the "Good" and "Bad" plots, there's no denying that the "Good" is better than the "Bad", but the key on the "Good" graph seems to me to commit one cardinal sin and also to miss an excellent opportunity. The sin is using squares in the key for markers that are all discs. The missed opportunity is that the markers could have been laid out as three points of the compass for East, South and West, and the central point for centre. I think that would make interpretation of the chart much more direct and much less error prone for many people.

On the central point (not using size to distinguish categories), however, I completely agree.

Mon 07 Feb 2011 at 10:39 PM

That Minard design is hanging in my cubicle. Though, what you've featured is a Tufte-redesign of the original ( which, I believe, is this: http://bit.ly/ha326L ) It's no less a great graphic, to be sure, but it's worth remembering that what I hung on my wall is a contemporized version.

Great stuff, overall. This is a fine collection of resources.

Tue 08 Feb 2011 at 5:14 AM

Depending on the audience and the point that was being made, I may have encoded size with profit ratio, or maybe gone with different markers for region, and color for profit ratio.

You could (and I hate adding complexity where not needed) add trend lines for each region.

Tue 08 Feb 2011 at 5:30 AM

I'd like to see this article note that while the horizon display may be simplistic, it is not a perfect solution.

Pilots untrained in instrumentation usually do not interpret the information correctly in hazardous conditions and sadly end up in a "graveyard spiral". The display is not made for the pilot, the pilot is trained to understand the display.

Tue 08 Feb 2011 at 11:14 AM

@ashley, thanks for your comments. In my presentation I said that the horizon graph is a very elegant way to present data in a small space. While it's obviously not ideal for finding exact values, it is good for getting a sense of extremes or rapid changes.

Additionally, my husband is an accomplished Black Hawk pilot as well as fixed wing pilot. He has over 20 years of experience flying. He reassures me that the attitude indicator was indeed made for pilots regardless of level of training. He also mentioned that novice pilots who are untrained in instrumentation are typically flying under VFR (visual flight rules).

Fri 11 Feb 2011 at 5:13 AM

My comments are based on Stanley Roscoe's work from 1972. He is credited with three basic principles, the most relevant in this case being the principle of the moving part. This principle states that moving part in a display should match our mental model of what moves. (Johnson & Roscoe, 1972, What moves, the airplane or the world?, Human Factors, 14, 107-129.)

In this case, our mental model of the world is that we are moving, not the horizon. So, it works against us for the horizon to move in the display. This can be improved slightly by giving motion to the airplane icon, but there's no denying that the elements are rooted in a way that works against the way we process information.

Your husband's comment about VFR is valid, but not all novice pilots find themselves in perfect conditions to use visual cues. One unfortunate example of this is JFK Jr.'s death. Link to article by Roscoe: http://www.aero.ca/e_kennedy's_spiral.html

From my studies, I understand that this is a debate that has been going on between the military and aviation psychologists since WWII. There are reasonable arguments on both sides; my goal is to simply share the knowledge from one side.

Wed 23 Feb 2011 at 4:40 AM
Commenting has been temporarily disabled.