Visual Communication

The New York Times Visualization Lab

2008-10-29T00:26:23-04:00

The New York Times' new Visualization Lab uses IBM's Many Eyes technology. While it provides easy access to a wealth of visualization techniques and the possibility to comment, there is one major difference: only data provided by the NY Times can be used. The kind and quality of that data will determine the success of this new site.

I criticized Many Eyes for not having a business model, but figured that they would be able to survive within an organization as large as IBM. Looks like they had a strategy, after all. Martin Wattenberg has also worked with the NY Times (he had a paper at this year's InfoVis conference), and he has an interest in InfoVis for the Masses.

The big difference between Many Eyes and the NY Times VizLab is that users cannot upload their own data. That means that the offered data will be crucial for the success of this site – if it's not interesting, people won't bother going there. And if the data is coming from online sources (and easy to obtain, like the data that is there right now), there will be little difference between the NY Times site and Many Eyes itself.

But that is where the New York Times can offer a huge value-add: by supplying data that cannot be easily found on the web, but that is collected by (or on behalf of) the NY Times. I'm specifically thinking of data like exit poll results, where usually only a small number of cross-sections are published. It would be excellent to have such data available to find some interesting comparisons of voters based on a number of criteria.

The NY Times name will certainly drive traffic, but to make the site compelling and make people come back, an investment in good data will be needed.

NY Times looks at Presidents and the Economy

2008-10-19T00:17:58-04:00

The New York Times has an interesting interactive visualization on the influence of presidents on the economy. They ask, Can a President Tame the Business Cycle? The visualization they use is not bad, but would be much more readable if it used a better color scale.

What exactly is a "high" or "low" change? This is how the legend describes the different colors used, and it turns out that "low" sometimes means negative. The color scale as shown in the legend is continuous, but one with just a few values (maybe five on either side of zero) would have been much more readable. Also, it is kind of important if things go up or down, which is impossible to see in this chart. Where exactly is zero on the color scale? The bar chart has no such problem.

(Click image for larger version)

The answer here is a diverging color scale with two colors that are different enough so that it is easy to see which side of zero a value is. ColorBrewer has a number of color scales for such (and other) purposes.

What is good about the graphic is its interactivity and the amount of data: almost 60 years of data, and seven dimensions is quite a bit of stuff to work with. There is also quite a bit of level-of-detail, with a mouse-over tooltip and a way to "drill in" for the bar/line chart.

Of course, it would be great if all of the data they collected for graphics like this were immediately available through their API ...

NY Times: The Best and Worst of Data Visualization

2008-09-24T00:04:01-04:00

The New York Times uses some of the best information graphics and visualizations on its web site and in the printed paper. But there is also a strange undercurrent of bad graphics, many of which commissioned from other sources, and often published in the New York Times Magazine. It almost feels like between all the good graphs, they need an outlet for the crazy stuff.

The NYTimes folks do not only know what's going on in information visualization: Matthew Ericson, thir Deputy Graphics Director, gave the keynote at InfoVis 2007. He talked about their attempts to adapt visualization techniques for use with a general audience and how they add hints for people to understand what is being shown.

They know how to use treemaps, and recently used one to show how the financial sector in the US has shrunk by 50% over the last year. And while this is a good visualization, their use of animation to provide transitions between the two points in time is very confusing. It is nearly impossible to follow what is going on. It takes me several seconds after each transition to find a particular institution again. This is how animation should not be done, and there is work out there that shows how to do this the right way.

I came across two other graphs in the NYTimes Magazine recently that also had me scratch my head. Looking through the graphs published this year, I found a few really good ones (including a good use of a 3D "brick"/"square pie" chart), a number of so-so ones, and a few really bad ones. These are clearly among the latter.

Exhibit A is from September 14, and looks at expectations of people to see a woman as President of the United States. These bubbles are very hard to compare, but what is more, they scaled the radius rather than the area. As we discussed recently, that has a quadratic effect on the area, exaggerating large numbers.

A bar chart may be boring (and also the usual response to a bad chart), but in this case a stacked bar chart would have made a lot more sense. Or even a number of (square) pie charts. Anything but this.

JunkCharts also recently discussed another bubble chart from the NYTimes, which has the opposite problem: All bubbles are (almost) the same size, even if the numbers are quite different. After these extremes, there's hope that there will be more correct bubble charts in the future, or even other kinds of charts that are easier to read to begin with, even without distortions.

Exhibit B is from September 21, and is an example of the ever popular pie chart, and perhaps the worst pie chart I have seen so far (and I've seen a few). Not only do the numbers not add up to 100%, the pseudo-3D makes the slices even harder to read than usual. The only thing the chart tells us is that there are different numbers of people who find that each of these professions prestigious. But a simple list would have accomplished the same thing. Making things more interesting is a good idea, but distorting the graph to the point of it being unreadable is not a price we should have to pay for that.

With great power comes great responsibility. The NYTimes is breaking new ground with their use of visualization and interaction, and I am grateful that they are really paying attention to what is happening in information visualization. But I wish they could reign in the their Department of Silly Charts and integrate it with the rest of the paper.

Engaging Readers with Square Pie/Waffle Charts

2008-09-07T22:12:03-04:00

Engaging viewers with interesting depictions of data always bears the risk of creating misleading or unreadable graphics. The square pie chart (or waffle chart) strikes a good balance between being interesting and not distorting the data. Here is an argument for the power of the pie and against the boredom of the bar.

The above chart was used in a series of articles on debt in the US in the New York Times. While not strictly a square pie chart, it does illustrate its main features: it's visually interesting, very readable and discoverable, and it does not distort the data.

Showing the same information in a bar chart may have been more "standard," but also rather dull. In addition, the differences between the values would have made it difficult to compare them. The matrix chart above lets the viewer easily work out how many times the average savings is owed in mortgages and other debt.

The chart also makes a point: that the savings are a tiny fraction of the debt. There is really no way not to see that. And that is visual communication beyond simple data visualization. A bar chart would not be able to do that, unless it essentially claimed that the savings were zero.

Kaiser at Junk Charts criticized the following chart (New York Times magazine, April 27, 2008):

He offered a bar chart as an alternative, which worked in this case because the values are not that far apart. But it was also as dull as any other chart, with nowhere near the visual interest of the "brick chart."

I actually think that this chart is quite clever. It uses a 10x10 grid as the base, so it is easy to read the numbers from looking at the number of layers. Despite the pseudo-3D and the fact that parts of the chars are occluded, the exact numbers can be read quickly. And the red brick metaphor even has a certain resemblance with school buildings.

In short, the chart has a style and a message. Good charts do not only have to be correct, they also have to be appealing. And this is well done here.

There is also a collection of good examples of pie charts in business graphics. Some of them may be a bit overloaded, but compared with other sins committed in business graphics and dashboards, it's very tame. I'd take a matrix/waffle/square pie chart over gauges and artificial horizons any day. Perhaps the influence of treemaps in business also helps make these charts look more familiar and thus more acceptable.

There is no doubt that we need to be careful about the choice of visual representation, and that we need to encourage the use of good charts and criticize the bad ones. But that doesn't mean we can get lazy and squeeze everything into a few standard charts types we've been using for decades. That is especially true if we want people to actually care about what we're trying to show – and not bore them to tears.

The Visual Display of Relevant Information

2008-07-21T20:43:20-04:00

When Al Gore talks about global warming, Hans Rosling shows the relationship between health and wealth, and the New York Times visualizes primary results and American consumer debt, they communicate visually. But they only use visual representation to get their point across, as a means to an end. When we want to show why visualization is effective, we have to care about the message, too – not just the method.

When pitching a tool or a project to a customer, one of the first tricks is showing them their own data. That is how they can relate to what the tool will do for them, rather than having to imagine how the presented toy data relates to what they care about. They see through the tool and see what they know. Or better yet, what they didn't know about their own data.

The same is true for showing visualization to a more general audience, where there is no "their data." But there is a lot of data that most people care about when it is presented to them, and that serves the same purpose. And when visualization is used to communicate not just the fact that the data exists, but also interesting and perhaps surprising information about it, people will listen (and watch).

Besides the examples at the very top of this posting, there are lots of others that may be less spectacular, but no less relevant. A recent project by Jeff Heer and colleagues at UC Berkeley together with Minnesota Public Radio looked at unemployment rates by sector in Minnesota over the last eight years. The brilliant Death and Taxes poster/interactive feature shows how the U.S. federal budget is split up between departments and programs, data every tax payer should be aware of (Jess Bachman, its designer, also has a very interesting blog where he looks at visualization problems like how to visualize the magnitude of one billion dollars).

Of course, visualization methods need to be developed by people who care about the visualization method, how it can be applied to different kinds of data, and how it supports different kinds of analysis and presentation.

But a compelling visualization needs a compelling story about interesting data. If it doesn't have that, it's no longer about effective visual communication. It becomes visualization porn.

Teaser image from the WallStats blog, used with permission.

For a similar problem in statistics, see "I don't care about the data ..."

The Unbearable Subjectivity of Visualization

2008-01-17T20:48:13-05:00

While reading Jarke van Wijk's Views on Visualization^(ref), I could not help but notice the negative references he makes to the subjectivity of visualization. A visualization science on par with statistics would certainly require the elimination of any and all subjectivity. I do not think that visualization is such a science, or that it being that is even all that desirable.

The subjectivity argument stands out especially in a paper titled Views on Visualization, which is more of a position statement than a traditional technical paper. While van Wijk presents ideas that are common in visualization (or at least acceptable once presented), he stresses the personal nature of his views in the introduction, and uses the first person deliberately when talking about his experiences and anecdotal evidence.

Subjectivity as a Problem

Clearly, subjectivity is something negative:

In the ideal case, one would hope that extraction of knowledge from data is an objective process, in the sense that the outcome does not depend on who performs it, and that the analysis can be repeated afterward by others, with the same outcome. Statistics aims at this, a typical pattern is the use of statistical tests to validate hypotheses on the data. Such tests make assumptions on the data (such as a normal distribution) and have free parameters (like the confidence level), but, furthermore, they do meet the criteria for objectiveness.
Unfortunately, visualization often does not meet this aim.

He then goes on to argue that it's not all bad, but gets very defensive: This does not mean that visualization is useless. And why would it?

It appears to me that we have the wrong ideas about visualization, and this is limiting our abilitiy to make visualization more than a manual alternative to data mining. Subjectivity is only bad if we suspect the user to twist the visualization to an extent where the visual result is not actually reflected in the data (which is undoubtably possible in many cases). But there are also other ways of looking at subjectivity, one of them is commonly known as the task. Statistics and data mining in all their objective glory cannot capture the intent of the user: what is the goal? What do I want to get out of the data? What does the data mean to me? There is of course an infinite number of objective truths about the any particular data set, and all but a very few are entirely meaningless to the user. Any actual analysis involves picking a particular subset of those truths to come to a conclusion, or to make a point.

Statistics is an obvious standard for comparison, especially for InfoVis (the paper is general enough to actually cover both Information and Scientific Visualization, though). Many techniques in InfoVis are based on statistical measures if not statistical graphics, and Martin Theus even claims that visualization is applied statistics – van Wijk seems to have a similar point of view. But why compete with statistics? And how? It's obvious that visualization will never have the accuracy a purely mathematical method has, or offer the thorough data digestion of data mining. Why go back to slide rules when we have pocket calculators?

Visualization: Careless Conduit or Complex Communication Conveyor?

It is easy to see that there is not just one right visualization for any given data set and task. And any visualization fulfills many more purposes than its designer foresees. We visualization designers are not simply applying our vast knowledge of visual representation, perception, and cognition to pipe data into the brain as efficiently as possible, but rather engage a complex and unpredictable mind. Our choices of parameters and interactions influence what the user sees and can do with a visualization. Our choice of data set to demonstrate a visualization technique prejudices the user's impression of how powerful the visualization can be. And the user's experiences, tasks, and goals.

A visualization is not simply a conduit for pumping information from the computer's hard disk into the user's brain. Such a view of visualization would only work if we were machines without the proper interfaces to directly plug into the data cloud. The power and beauty of visualization goes much further, and provides many interesting challenges along the way.

Objective Visualization: How?

Let's try a little gedankenexperiment to see how viable an objective visualization would be. In addition to the perfect visualization method we would need the proper conditions: a calibrated display to get proper color reproduction, controlled ambient light that is not distracting and not influencing the user's perception, an environment that is free from shapes that could prejudice the user's mind by being recognized in the visualization, etc. Further, the user would need to be standardized: properly trained in visualization as well as the particular data set, with no pre-existing ideas about the data, no interest in a specific outcome of the analysis, and clearly no color-blindness, color deficiencies, less-than-perfect vision, or other perceptual or cognitive deviations from the norm.

Even if all that was possible, would that be desirable? What would be the point? Just as visualization can be and is used in subjective ways, objectivity of statistics is only skin-deep: choosing the right measures and parameters is subjective and task-dependent! The entire point of letting humans deal with data sets is that there is value in subjectivity, experience, and opinion – plus the fact that the the consumers of an analysis are human, too!

Visualization is Communication

This is what is commonly known as expression and communication. Assuming there has been no downright lying, we have to accept any interpretation of the data as valid, just as we have to accept any statistical measure. But by emphasizing different aspects of the data, by picking subsets, or by fiddling with settings, we can present different points of view. That may make things more complex, but also more interesting.

Does that mean that anything goes? Any visualization is good and useful? Of course not, but we need to broaden our ideas and criteria for good visualizations. There is never only one truth about a data set, regardless of how it is analyzed or presented. This was a major point of discussion at the Social Data Visualization Panel, and perhaps the crucial one. Academics can't control visualization, nobody will ask us to sanction every visualization that is produced. We need to learn to let go.

Conclusions

I don't disagree with van Wijk's points about subjectivity, but I think we are doing ourselves a disservice by trying to eliminate, control, or even just downplay it. Subjectivity is a great strength that may be daunting to the technically minded person, but that makes things like photography, email, language, etc. so powerful. Just like there is no central authority for which truths can be expressed in English, there is no central authority for visualization. Why would we want things to be different?

2007 was the year of social visualization. With any luck, 2008 will be when people will start embracing the subjective nature of visualization and realize that visualization is communication.

Update: Jorge Camoes has posted an interesting discussion of Minard's Napoleon's March that provides food for thought along similar lines.

Image credits: Left half taken from the cited paper; right part from illustration for the Wikipedia entry on communication by Luis Javier Rodriguez Lopez.

Expressive Visualization, Updated Presidents Chart

2007-11-30T00:29:17-05:00

I used the Presidential Demographics chart in my talk at the Impact of Social Data Visualization panel at InfoVis 2007, and got some interesting responses to that. There is some interest in printing this out, so I have made a new version of the chart that is now also available as a PDF. Stephen Few used Joseph Berk's term "interocular traumatic impact" – a visualization that hits you between the eyes – to describe it. And this is exactly what visualization can do extremely effectively: visual communication, and not just of data.

Visualization can tell a story, just like a comic/graphic novel or a silent movie can. Expressive visualization takes the idea of visualization as visual communication further, and uses almost exclusively visual means to make a point. And it doesn't even have to be dumbed down or prettied up for that purpose: a simple chart can be much more effective in delivering such a message.

Most of us have probably seen cases where a visualization made one point stand out, even if that may not have been the most important one. We need to learn to control the strength of a visualization, and make it work for us.

I am interested in finding more examples of visualizations that make a point – political or not. Any pointers would be appreciated.