Tuesday, September 25, 2007

the semantic side of visualization

When I started thinking about how to analyze the design of information visualization, and the ways in which it could be made more "layperson-centric," my initial inclination was to focus exclusively on its visual syntax as something completely separate from its semantic content. In terms of programming design patterns, this would be the Model-View-Controller approach, where the visualization itself represents the "view" and "controller," and its construction ignores the internal details of the "model" (or, in this case, the data or information being visualized). I think there are a number of valid reasons to think in these terms (one obvious one being that "make visualizations about more interesting data" is not a particularly effective visualization design suggestion), but I've also come to realize two things: One, that visual design decisions can be greatly informed by the semantic content they address. Two, that visualization effectiveness in general can be enhanced by "extra-visual" considerations, such as the context in which they are delivered. A fairly trivial example of the first would be the use of maps to relate census data; the visuals produced by WorldMapper.org carry an impact and effectiveness that would not necessarily come through in a graph or chart representation. The following are a couple of examples of the second that I have been thinking about recently.

Last week, the MIT Communications Forum (hosted by the Comparative Media Studies department) held its first event of the Fall, a forum entitled "What is Civic Media?" (podcast available here). As a sort of launch-party for the new MIT Center for Future Civic Media (a collaboration between CMS and the Media Lab), the forum addressed the new possibilities for civic engagement and "participatory democracy" afforded by the rise of "Web 2.0 related" media (blogs, *casts, virtual worlds, etc.). To my pleasant surprise, information visualization was well represented as an important tool for promoting these practices! One of the panelists, Ethan Zuckerman of Harvard's Berkman Center for Internet and Society, presented several map-based projects designed to visualize media attention across the globe. Beth Noveck of NYU Law School explicitly stated the need for effective visualization of civic data to engage and inform participant citizens. I couldn't agree more!

She mentioned several existing tools out there, including the sense.us project (more on this in a minute), but I was most intrigued by her reference to smartvote.ch, a site that has evidently been deployed in Switzerland for several years now. In their words:

smartvote is a scientifically conceived online election aid for communal, cantonal and national elections in Switzerland.

smartvote aims to:

  • improve the transparency of elections and give voters a new way to make an
    informed choice

  • increase people's interest in politics

  • demonstrate the potential of e-democracy and e-voting

The political profiles of candidates are established through questions about
their policies and attitudes, and these are saved in a database. Voters can
then answer the same questions, and smartvote will recommend to them those candidates who show the closest political match.

(Please excuse any messy formatting -- I've about had it with Blogger's post editor!)

The site has a strong visual component, where candidate's positions on various issues (gleaned from a questionnaire) are presented in several graphical formats, including a spider chart that creates a representation of their political "footprint." What makes it more compelling, though, is that a user can fill out the questionnaire as well, and see how their spider chart compares with those of the various candidates. This may be a trivial observation, but this sort of interactivity (at the "extra-visual" level) undoubtedly makes the system more engaging. Essentially they are offering the user the ability to insert themselves in to the data being visualized; we can easily imagine that a user would be more compelled to explicitly match candidate "footprints" to their own rather than simply browsing the footprints with only an abstract idea of what they are looking for. A huge part of engaging a user in a visualization tool (or any tool, for that matter) is getting them invested in what they are looking at; what better way than allowing them to see themselves in the data (incidentally, this may have been the sort of entry-point that made the Baby Name Voyager so popular)? The smartvote.ch site has only limited English translation, so I'd be curious to find out how it is being used and received by its audience.

The second visualization project using extra-visual techniques to great effect is the aforementioned sense.us and, more recently, Many Eyes. Developed by the folks at the IBM Visual Communications Lab, including Fernanda Viegas and Martin Wattenberg, both of these tools emphasize "the social side of visualization." In sense.us (which doesn't appear to have a live demo anymore), users can explore a series of visualizations of United States census data. While the visuals themselves are quite nice and clear, the key feature of the tool is the way that it encourages collaborative analysis, "recasting visualizations as not just analytic tools, but social spaces." Among other features, sense.us embeds comment boards within the visualization, allowing users to comment on, and mark up, particular features of the graphs (including the ability to "hyperlink" to specific views of the data, so that other users can see exactly what you have discovered through your manipulation of filters, etc.). This is all designed, obviously, to encourage discussion of the data -- another great way to engage and invest users. (Martin and Fernanda gave a great presentation on Many Eyes last Spring at the CMS graduate colloquium. For some reason the podcast of the talk was never posted, but I'm working on rectifying this and will post a link to it in the near future.)

This feature set continues in Many Eyes, which is a live service that allows users to upload their own data sets and create visualizations of them based on a series of graph and chart templates provided by the site. Once again, the graphics are quite nice (and all fully interactive), but the real draw is the "extra-visual" social element that encourages discussion and collaboration. I see it as a way of subverting the sorts of purely visual issues that come up when designing information visualization tools (some of which I have written about here already); essentially, it recognizes that a universally readable visual syntax may be impossible to produce, and simply uses this social framework as a "helper." Aside from more analytical discussion, a user confused or intimidated by the data has the ability to voice that confusion and learn from other users.

The value of this kind of interaction shouldn't be overlooked - I saw this while designing educational visualizations. Visualization can be imposing and confusing; many times it presents itself as a stand-alone, fully functional tool that dispenses information and expects to be understood without much qualification. It's very easy for users to become discouraged in this situation if they can't immediately "read" what is being presented (Donald Norman talks about this in "The Design of Everyday Things"), which is exactly the opposite of the experience someone should have with information visualization. Incorporating this social framework, I think, goes a long way towards easing this tension. If nothing else, it reminds users that visualization is for the people.

The final project I've been interested in along these lines is the work Hans Rosling has done at gapminder.org. Rosling has been using the Trendalyzer software to develop interactive, animated tools that "visualize human development." I first discovered him through his two great TED talks ("New insights on poverty and life around the world," and "Debunking third-world myths with the best stats you've ever seen"), which were extremely entertaining (for me, at least!). However, while the software itself seems well designed, what makes his presentations so compelling is the narrative that he builds around the software and the data it is presenting. Rosling acts as a sports announcer, giving a running commentary on the graphics as they move across the screen, and not only is his excitement for the data contagious, but it clearly increases the effectiveness of the visualizations (evidently so much so that the new "gapcasts" at gapminder.org feature Rosling green-screened over the Trendalyzer images, continuing to narrate the action!).

So, what does this mean in terms of visualization design? In this case it isn't totally clear to me yet. Obviously, information visualization should be a narrative, and Rosling has expertly produced examples of this. But how might designers incorporate some of the "magic" that Rosling generates in these presentations without the services of a charismatic Swedish sword-swallower (watch the first TED talk all the way to the end!)? In principle it should be possible, as he isn't cheating too much; most of the information he references in his narration is contained within the visualization he is presenting. So is it just his charismatic performance that adds value here? Certainly that's part of it, but ultimately what he's doing is serving as a guide, providing a guided tour of the data. In this sense, he's doing something that could conceivably be implemented in other ways, which might mean providing context for the visualization, or "tutorials" for its functionality, or finding other ways of increasing its narrative qualities. Or perhaps this overlaps with the sort of social activity used in Many Eyes. Or maybe information visualization does need more Swedish sword-swallowers.

In any case, what these examples indicate to me is that visualization tools are more than just visuals. While pure visual design works in many cases, addressing semantic and "extra-visual" content, directly or indirectly, can also encourage interaction and investment.

Tuesday, September 11, 2007

information and aesthetics

One of the dimensions of information visualization that I’m particularly interested in is the impact of aesthetics on the design of functional visualization tools “for the people.” This is basically the age-old debate of form versus function, or beauty and expressiveness versus practicality and usability, or any of the variations on that theme. As someone who was trained in both visual art and physics, I’ve spent a lot of time thinking about this (ostensible) opposition in various contexts over the years (as anecdotal evidence, I consider most of my work here to be “aesthetic remixes” of versions I developed for more formal educational purposes). It’s also a fairly subtle one for a number of reasons, but particularly because the “first principles” of aesthetics aren’t as easily categorized as those of other fields that inform information visualization design (graphic design, perception/cognition, semiotics, HCI, etc.), making it difficult to establish what works and what doesn’t.

Last week I criticized Twitter Blocks on the grounds that it was lacking in usability, despite being aesthetically compelling. Tom Carden’s response (as one of the designers of Twitter Blocks) got me thinking more about the role of aesthetics in visualization once again. While I understand and agree with his position completely (visualizations need not necessarily be useful to be interesting), I’m mostly concerned with the “traditional” goals of information visualization, and the ways in which we can develop a more robust literacy for them. In that sense, I’m inclined to promote usability over aesthetic novelty. That said, what’s clear to me from examples out “in the wild” like Twitter Blocks and the DiggLabs stuff (I don’t mean to pick on Stamen’s work, it’s just that they’re one of the most ubiquitous popular visualization designers out there right now), and the discussion surrounding them, is that the aesthetics are compelling: In walking the line Tom describes in his post, Stamen has produced a body of visualization work that both excites (“Beautiful!”) and frustrates (“Useless!”). This suggests that incorporating an aesthetic approach to visualization design can be beneficial in its capability to engage users, but that taking it “too far” (whatever that means in specific cases) may be counter-productive if we are looking to develop a type of standard visualization literacy. The big and difficult question, then, is determining what aesthetic considerations we can take away from these and other examples that will support this goal.

I recently discovered a set of research papers by Adrew Vande Moere (of infosthetics fame) and his grad students at the University of Syndey, Nick Cawthon and Andrea Lau, that attempt to address this question:

The first begins to describe the role of aesthetics in visualization, arguing that it is as important an influence on user experience as the more traditional components of information visualization. Parallels are drawn to the aesthetics of cinematography, choreography and staging as a means to streamline a visual presentation. An appeal is made to the importance of aesthetics in producing effective information narratives, in developing useful visual metaphors, and in using interaction to encourage “play.” I agree with all of these points, and I think they are particularly relevant to the concept of “information visualization for the people.” Part of what makes a visualization successful (and, presumably, effective) is its ability to engage the user, and I’m certain these techniques, in principle, promote engagement. I spent a lot of time thinking about designing playful tools when I was working on educational visualizations and simulations for teaching physics and biology. The trick was to make them forget they were learning!

That said, while I think this is the right way to frame an analysis of aesthetics in visualization, the problem in its application is usually one of degree. The informed “use of aesthetics” (a difficult to define concept) can go a long way towards improving effectiveness, but without a measured approach, you run the risk of interfering with the effective presentation of your data. Ignoring the designer’s intent, this is my critique of many of the more popular visualizations on the web right now: too great a focus on beauty and not enough on functionality. This is fine for work intended to be artistic, but it defies the development of a more functional visualization literacy.

The second paper addresses this issue by categorizing various “sub-genres” of visualization along an aesthetic spectrum:

I was less convinced by this analysis, for a couple of reasons: First, it seemed more concerned with promoting “information aesthetics” as a valid visualization field (that “reaches beyond” information visualization) than with providing concrete details of what its benefits actually are. It establishes a large “grey area” to be defined as information aesthetics, but left me wondering about the nuances within that space. Second, I’m not entirely convinced that it makes sense to divide visualization-space in to these vague categories; I’m more inclined to divide the space in to two broad categories (information visualization, and “information-based art”), rather than confusing the taxonomy with all these sub-genres (that sort of categorization reminds me of the taxonomy of electronic music). Some of the rhetoric in the paper suggests that designers decide to produce work within these sub-genres, rather than the work being assigned to a genre by virtue of its characteristics. If I’m being very cynical, it almost sounds like a broad space is being carved out to support information design failures; if an information visualization isn’t coherent enough to be effective, don’t worry – it was supposed to be an information aesthetic visualization.

The third paper was the most interesting. As the title suggests, it presents research aimed at quantifying the effects of aesthetics on visualization usability. The researchers devised a web-based survey that asked participants to answer information-retrieval questions based on a series of visualizations characterized by (presumably) varying levels of aesthetic attractiveness. They were primarily measuring response time; in particular, erroneous response time (the amount of time spent on a question before getting it wrong) and latency of task abandonment (the amount of time a user spends on a question before opting to “give up”). Their theory was that increased aesthetic quality would decrease these times. Without going in to the details, they were right. The visualization deemed most attractive by the participants generally performed best in the survey. Very interesting to see this quantified in some way.

However, to be fair, I felt there were several issues with their survey (which I took myself the other day), most of which are recognized in the paper. The most obvious had to do with the fact that the visualizations used in the survey were all static, non-interactive images. A few of them were snapshots from three dimensional systems, which really require interactivity to make sense of them (by allowing users to rotate the structures, etc.), so those got a raw deal. But even for the 2D ones, viewing static images isn’t representative of information visualization. Similarly, despite the fact that the test visualizations were representatives of actual visualization systems, they didn’t strike me as representative of the sorts of visualizations (information, or information aesthetic) that are out in the wild these days. These were pretty conservative as far as aesthetic detail goes, and the use of a consistent color palette reinforced this. Oddly enough, this choice actually struck me as representative of the sorts of “standardization” I’ve been promoting to encourage popular literacy. I would be really interested to see a similar survey involving actual examples from the web. The measure of aesthetic is qualitative enough that enforcing consistency between survey examples seems unnecessary and maybe counter-productive.

Another possibly significant issue was the participant demographic. It wasn’t a randomly selected sample – the survey was primarily advertised to people in the design community, who most likely have a higher than average visual literacy. If we’re trying to determine the effects of aesthetics on visualization usability, these probably aren’t the best people to survey. Also interesting to note was that nearly half of the participants in the survey did not complete it; their data wasn’t used in the results.

Finally, I was slightly amused by what almost sounds like an admission of bias in their conclusion: “The original purpose of this research was to increase the awareness of the positive role and purpose of aesthetic in the design of data visualization techniques.” Does this mean anything? Who knows!

In any case, these were all great reads. They’re not long, so I encourage anyone interested in these questions to look them over. While I’m not totally convinced by some of their conclusions, I’m glad that someone is asking them. Aesthetics can undoubtedly inform a visualization literacy, but nailing down how much is too much is definitely tricky.

Friday, September 7, 2007

same old problems

While I'm mostly interested in more complex interactive visualization tools, you can't really talk about a literacy for them without considering our literacy for simple, static charts and graphs. I recently came across a (kind of old) post about these graphics on Mike Dickison's Pictures of Numbers blog ("Deceptive Areas," "False Advertising"). They appeared in the New York Times Magazine a few months ago. As Dickison points out in his analysis, the number of inconsistencies and flat-out errors in these charts is almost comical (my personal favorite is in "The Ad Buyers" graph, where the line for item 2 starts above the line for item 3, despite that its value, 3.1, is less than the value of item 3, 3.8). These are the sorts of mistakes Edward Tufte was talking about 30 years ago, and yet we're still seeing them in the 2K7!
Obviously I don't want to make generalizations from a couple of bad examples (although I'm sure these aren't isolated incidents), but what does this mean? Clearly the designers themselves aren't paying attention to issues involved in the graphical presentation of information. The editors (of an ostensibly respectable publication) aren't catching these issues. Finally, the repeat offenses suggest that readers aren't complaining either. This seems like a pretty dismal situation.
I've read that those types of graphics started being added to stories to draw in readers who otherwise wouldn't be bothered to read the article. Does this mean that we generally don't pay attention to info-graphics anymore either? That we see a chart or graph and just skip it? Or that we are satisfied with a pretty design? What are the implications of these sorts of situations to our visual literacy?
This is something I would like to come back to again sometime.

Thursday, September 6, 2007

twitter blocks

I caught this story yesterday on Matthew Hurst's data mining blog, along with more discussion at TechCrunch, and wanted to write something about it while it was still hot.

"Twitter Blocks" is a new visualization of data from the social networking (or whatever they're calling it) site, Twitter. As I understand it, Twitter allows users to post text-message sized nuggets of information about what they're up to at any particular moment. The idea is to post these short messages relatively frequently (from your phone, even) in order to build up a timeline of updates that describes your life. As with any good social networking site, you can explore what other Twitter users are posting, add them as friends, etc.. The "Twitter Blocks" tool, developed by Stamen Design (of Digg.com visualization fame; more on these in a future post), is an attempt to visually display your local network of Twitter friends. Unfortunately, while I think that Blocks is aesthetically attractive, and I applaud Stamen for being prolific visualization producers, as a functional information visualization it is pretty weak.

Granted, I do not use Twitter and have only a vague idea of how it works, but even after studying Blocks for 20 minutes, I have absolutely no idea what metrics it is representing. Here is my short list of issues with it:

  • Lack of identifying labels (aside from the cryptic "public timeline ->"). The layout suggests a chart or graph or timeline, but there is no indication of what is being "measured" or what metrics its dimensionality maps to.
  • As is typical of Stamen's designs (see Digg visualizations), the "help" screen offers very little help. The single sentence description ("Blocks are recent statuses from users and the people they follow") at best gives no further insight in to the design. For someone not familiar with Twitter it's downright nonsensical.
  • Some "blocks" extend upwards or downwards off the screen; what does this mean? Again, what does the height signify at all, for that matter?
  • The three dimensional structure of the chart is difficult to discern because the viewpoint only rotates a few degrees from its default position, preventing you from getting a clearer view of parts of the "graph."
  • The "zoom" feature (when you click on a block) happens too quickly to understand what's happening.
  • In some situations (see screenshot above) the chart takes on Escher-esque features, with "blocks" that are further away being drawn on top of closer ones (although I assume this is a bug).

By the standards of cannonical information visualization, this is a sloppy design. But that begs the question: how does the question of form versus function apply to tools like Twitter Blocks? Is Twitter Blocks intentionally not useful? I thought the discussion at TechCrunch was interesting because I rarely see users weighing in on this question. Particularly, Tom Carden, a designer at Stamen (and someone whose work I have enjoyed in the Processing community) posted an interesting comment:

"bankblast - plenty of things are cool and useless if you stop to think about it (TV? movies? games?). So I don’t have a problem with that. But we’ve been using Blocks in the office for a couple of weeks now and I keep finding out new things that I couldn’t have seen on my Twitter homepage. Worksforme!"

No offense to Tom, but it sounds like he's playing both sides! He admits to its potential uselessness but simultaneously suggests that that it is quite functional. Based on Stamen's previous work, I do think they are trying to produce useful information tools and not just pretty designs, but justifications like this seem like an easy way out of more careful consideration of their design. And frankly, the "works for me" defense seems completely antithetical to the principles of information visualization!

Anyways, the question of form versus function in information visualization is one that I am really interested in. I hope to write more about this in the future.

Wednesday, September 5, 2007

first p0st~!!1

This is the first post on my awesome new blog!!

My name is Mike Danziger, I am a graduate student in Comparative Media Studies at MIT, and my intent is to use this blog as a platform for developing and researching ideas associated with my thesis. Mostly I want to encourage myself to get writing on the subject, but, in the best case scenario, I'm also interested in generating some discussion around these topics. Assuming anyone ever reads this, the blog format seems well suited to these goals.

That said, my research interest is in information visualization. In particular, I'm interested in what I perceive as its imminent (if not already underway) transition from a relatively niche research field to a "mainstream" communicative medium. Up until now, the field has arguably seen application primarily within "special interest" groups dealing in large data sets: scientific fields needing to display experimental data, medical fields interested in the imaging of test results, and business and military organizations tracking large data trends (stock market, network traffic, etc.). While information visualization has succeeded here, its application has been largely characterized as "by experts, for experts." The users of these visualization tools were presumably highly trained, and the tools were designed, perhaps inadvertently, with that in mind.

Today's visualization needs are different. Without resorting to too much "information super-highway" rhetoric, the internets have put almost limitless information at the fingertips of anyone with a network connection, and navigating this information deluge has become an integral part of our everyday lives. This has been most recently emphasized by the "Web 2.0" movement, where organizations are beginning to deliver raw information directly to their users, and allowing them to interact with it on their own terms (ranging from services like Amazon.com and Google to social networks like Facebook and MySpace). With this increased access to information comes an increasing need for efficient ways of attending to it. Logically, the need for information visualization tools is greater than ever, but today's information consumers ("the people") are not necessarily experts at using them.

I want to suggest that information visualization, as a field, is having some trouble making the transition to a form that is understandable to the layperson. It's not necessarily for lack of trying -- increasingly high-profile examples of "popular" visualization are appearing all over the web these days -- but more likely the result of a wide array of competing (or interfering) assumptions informing the design of visualizations "for the people." The evidence of this is plainly visible in a survey of examples of "popular" information visualization across the web (see some of the related links on the side of the page); there is a huge variety of visual (and conceptual) constructions being employed in their design, and often very few commonalities, even when presenting similar types of information. While I am a believer in variety being the spice of life, I would argue that this type of schizophrenic approach hinders our ability to develop a general literacy for information visualization. If information visualization is going to become mainstream, a "visualization literacy" is exactly what is needed.

I've only vaguely gestured at some ideas here, but my goal with this blog is to explore some of the assumptions underlying "popular" visualization design and reception, in the hopes of uncovering some practical, unifying principles that could be applied to future design. Or, at the very least, draw attention to some of the existing design issues in the hopes that awareness and discussion of them will result in more informed design. Either way, I'm interested in discovering ways in which we can promote a common literacy for visualization, thus ushering in a brave new era of visually-enhanced information-rich existence!!! To that end, I hope anyone reading this -- designer, consumer, or otherwise -- will feel free to contribute.